fs/ocfs2/namei.c

   1 // SPDX-License-Identifier: GPL-2.0-or-later
   2 /* -*- mode: c; c-basic-offset: 8; -*-
   3  * vim: noexpandtab sw=8 ts=8 sts=0:
   4  *
   5  * namei.c
   6  *
   7  * Create and rename file, directory, symlinks
   8  *
   9  * Copyright (C) 2002, 2004 Oracle.  All rights reserved.
  10  *
  11  *  Portions of this code from linux/fs/ext3/dir.c
  12  *
  13  *  Copyright (C) 1992, 1993, 1994, 1995
  14  *  Remy Card (card@masi.ibp.fr)
  15  *  Laboratoire MASI - Institut Blaise pascal
  16  *  Universite Pierre et Marie Curie (Paris VI)
  17  *
  18  *   from
  19  *
  20  *   linux/fs/minix/dir.c
  21  *
  22  *   Copyright (C) 1991, 1992 Linux Torvalds
  23  */
  24
  25 #include <linux/fs.h>
  26 #include <linux/types.h>
  27 #include <linux/slab.h>
  28 #include <linux/highmem.h>
  29 #include <linux/quotaops.h>
  30 #include <linux/iversion.h>
  31
  32 #include <cluster/masklog.h>
  33
  34 #include "ocfs2.h"
  35
  36 #include "alloc.h"
  37 #include "dcache.h"
  38 #include "dir.h"
  39 #include "dlmglue.h"
  40 #include "extent_map.h"
  41 #include "file.h"
  42 #include "inode.h"
  43 #include "journal.h"
  44 #include "namei.h"
  45 #include "suballoc.h"
  46 #include "super.h"
  47 #include "symlink.h"
  48 #include "sysfile.h"
  49 #include "uptodate.h"
  50 #include "xattr.h"
  51 #include "acl.h"
  52 #include "ocfs2_trace.h"
  53
  54 #include "buffer_head_io.h"
  55
  56 static int ocfs2_mknod_locked(struct ocfs2_super *osb,
  57                               struct inode *dir,
  58                               struct inode *inode,
  59                               dev_t dev,
  60                               struct buffer_head **new_fe_bh,
  61                               struct buffer_head *parent_fe_bh,
  62                               handle_t *handle,
  63                               struct ocfs2_alloc_context *inode_ac);
  64
  65 static int ocfs2_prepare_orphan_dir(struct ocfs2_super *osb,
  66                                     struct inode **ret_orphan_dir,
  67                                     u64 blkno,
  68                                     char *name,
  69                                     struct ocfs2_dir_lookup_result *lookup,
  70                                     bool dio);
  71
  72 static int ocfs2_orphan_add(struct ocfs2_super *osb,
  73                             handle_t *handle,
  74                             struct inode *inode,
  75                             struct buffer_head *fe_bh,
  76                             char *name,
  77                             struct ocfs2_dir_lookup_result *lookup,
  78                             struct inode *orphan_dir_inode,
  79                             bool dio);
  80
  81 static int ocfs2_create_symlink_data(struct ocfs2_super *osb,
  82                                      handle_t *handle,
  83                                      struct inode *inode,
  84                                      const char *symname);
  85
  86 static int ocfs2_double_lock(struct ocfs2_super *osb,
  87                              struct buffer_head **bh1,
  88                              struct inode *inode1,
  89                              struct buffer_head **bh2,
  90                              struct inode *inode2,
  91                              int rename);
  92
  93 static void ocfs2_double_unlock(struct inode *inode1, struct inode *inode2);
  94 /* An orphan dir name is an 8 byte value, printed as a hex string */
  95 #define OCFS2_ORPHAN_NAMELEN ((int)(2 * sizeof(u64)))
  96
  97 static struct dentry *ocfs2_lookup(struct inode *dir, struct dentry *dentry,
  98                                    unsigned int flags)
  99 {
 100         int status;
 101         u64 blkno;
 102         struct inode *inode = NULL;
 103         struct dentry *ret;
 104         struct ocfs2_inode_info *oi;
 105
 106         trace_ocfs2_lookup(dir, dentry, dentry->d_name.len,
 107                            dentry->d_name.name,
 108                            (unsigned long long)OCFS2_I(dir)->ip_blkno, 0);
 109
 110         if (dentry->d_name.len > OCFS2_MAX_FILENAME_LEN) {
 111                 ret = ERR_PTR(-ENAMETOOLONG);
 112                 goto bail;
 113         }
 114
 115         status = ocfs2_inode_lock_nested(dir, NULL, 0, OI_LS_PARENT);
 116         if (status < 0) {
 117                 if (status != -ENOENT)
 118                         mlog_errno(status);
 119                 ret = ERR_PTR(status);
 120                 goto bail;
 121         }
 122
 123         status = ocfs2_lookup_ino_from_name(dir, dentry->d_name.name,
 124                                             dentry->d_name.len, &blkno);
 125         if (status < 0)
 126                 goto bail_add;
 127
 128         inode = ocfs2_iget(OCFS2_SB(dir->i_sb), blkno, 0, 0);
 129         if (IS_ERR(inode)) {
 130                 ret = ERR_PTR(-EACCES);
 131                 goto bail_unlock;
 132         }
 133
 134         oi = OCFS2_I(inode);
 135         /* Clear any orphaned state... If we were able to look up the
 136          * inode from a directory, it certainly can't be orphaned. We
 137          * might have the bad state from a node which intended to
 138          * orphan this inode but crashed before it could commit the
 139          * unlink. */
 140         spin_lock(&oi->ip_lock);
 141         oi->ip_flags &= ~OCFS2_INODE_MAYBE_ORPHANED;
 142         spin_unlock(&oi->ip_lock);
 143
 144 bail_add:
 145         ret = d_splice_alias(inode, dentry);
 146
 147         if (inode) {
 148                 /*
 149                  * If d_splice_alias() finds a DCACHE_DISCONNECTED
 150                  * dentry, it will d_move() it on top of ourse. The
 151                  * return value will indicate this however, so in
 152                  * those cases, we switch them around for the locking
 153                  * code.
 154                  *
 155                  * NOTE: This dentry already has ->d_op set from
 156                  * ocfs2_get_parent() and ocfs2_get_dentry()
 157                  */
 158                 if (!IS_ERR_OR_NULL(ret))
 159                         dentry = ret;
 160
 161                 status = ocfs2_dentry_attach_lock(dentry, inode,
 162                                                   OCFS2_I(dir)->ip_blkno);
 163                 if (status) {
 164                         mlog_errno(status);
 165                         ret = ERR_PTR(status);
 166                         goto bail_unlock;
 167                 }
 168         } else
 169                 ocfs2_dentry_attach_gen(dentry);
 170
 171 bail_unlock:
 172         /* Don't drop the cluster lock until *after* the d_add --
 173          * unlink on another node will message us to remove that
 174          * dentry under this lock so otherwise we can race this with
 175          * the downconvert thread and have a stale dentry. */
 176         ocfs2_inode_unlock(dir, 0);
 177
 178 bail:
 179
 180         trace_ocfs2_lookup_ret(ret);
 181
 182         return ret;
 183 }
 184
 185 static struct inode *ocfs2_get_init_inode(struct inode *dir, umode_t mode)
 186 {
 187         struct inode *inode;
 188         int status;
 189
 190         inode = new_inode(dir->i_sb);
 191         if (!inode) {
 192                 mlog(ML_ERROR, "new_inode failed!\n");
 193                 return ERR_PTR(-ENOMEM);
 194         }
 195
 196         /* populate as many fields early on as possible - many of
 197          * these are used by the support functions here and in
 198          * callers. */
 199         if (S_ISDIR(mode))
 200                 set_nlink(inode, 2);
 201         mode = mode_strip_sgid(dir, mode);
 202         inode_init_owner(inode, dir, mode);
 203         status = dquot_initialize(inode);
 204         if (status)
 205                 return ERR_PTR(status);
 206
 207         return inode;
 208 }
 209
 210 static void ocfs2_cleanup_add_entry_failure(struct ocfs2_super *osb,
 211                 struct dentry *dentry, struct inode *inode)
 212 {
 213         struct ocfs2_dentry_lock *dl = dentry->d_fsdata;
 214
 215         ocfs2_simple_drop_lockres(osb, &dl->dl_lockres);
 216         ocfs2_lock_res_free(&dl->dl_lockres);
 217         BUG_ON(dl->dl_count != 1);
 218         spin_lock(&dentry_attach_lock);
 219         dentry->d_fsdata = NULL;
 220         spin_unlock(&dentry_attach_lock);
 221         kfree(dl);
 222         iput(inode);
 223 }
 224
 225 static int ocfs2_mknod(struct inode *dir,
 226                        struct dentry *dentry,
 227                        umode_t mode,
 228                        dev_t dev)
 229 {
 230         int status = 0;
 231         struct buffer_head *parent_fe_bh = NULL;
 232         handle_t *handle = NULL;
 233         struct ocfs2_super *osb;
 234         struct ocfs2_dinode *dirfe;
 235         struct ocfs2_dinode *fe = NULL;
 236         struct buffer_head *new_fe_bh = NULL;
 237         struct inode *inode = NULL;
 238         struct ocfs2_alloc_context *inode_ac = NULL;
 239         struct ocfs2_alloc_context *data_ac = NULL;
 240         struct ocfs2_alloc_context *meta_ac = NULL;
 241         int want_clusters = 0;
 242         int want_meta = 0;
 243         int xattr_credits = 0;
 244         struct ocfs2_security_xattr_info si = {
 245                 .name = NULL,
 246                 .enable = 1,
 247         };
 248         int did_quota_inode = 0;
 249         struct ocfs2_dir_lookup_result lookup = { NULL, };
 250         sigset_t oldset;
 251         int did_block_signals = 0;
 252         struct ocfs2_dentry_lock *dl = NULL;
 253
 254         trace_ocfs2_mknod(dir, dentry, dentry->d_name.len, dentry->d_name.name,
 255                           (unsigned long long)OCFS2_I(dir)->ip_blkno,
 256                           (unsigned long)dev, mode);
 257
 258         status = dquot_initialize(dir);
 259         if (status) {
 260                 mlog_errno(status);
 261                 return status;
 262         }
 263
 264         /* get our super block */
 265         osb = OCFS2_SB(dir->i_sb);
 266
 267         status = ocfs2_inode_lock(dir, &parent_fe_bh, 1);
 268         if (status < 0) {
 269                 if (status != -ENOENT)
 270                         mlog_errno(status);
 271                 return status;
 272         }
 273
 274         if (S_ISDIR(mode) && (dir->i_nlink >= ocfs2_link_max(osb))) {
 275                 status = -EMLINK;
 276                 goto leave;
 277         }
 278
 279         dirfe = (struct ocfs2_dinode *) parent_fe_bh->b_data;
 280         if (!ocfs2_read_links_count(dirfe)) {
 281                 /* can't make a file in a deleted directory. */
 282                 status = -ENOENT;
 283                 goto leave;
 284         }
 285
 286         status = ocfs2_check_dir_for_entry(dir, dentry->d_name.name,
 287                                            dentry->d_name.len);
 288         if (status)
 289                 goto leave;
 290
 291         /* get a spot inside the dir. */
 292         status = ocfs2_prepare_dir_for_insert(osb, dir, parent_fe_bh,
 293                                               dentry->d_name.name,
 294                                               dentry->d_name.len, &lookup);
 295         if (status < 0) {
 296                 mlog_errno(status);
 297                 goto leave;
 298         }
 299
 300         /* reserve an inode spot */
 301         status = ocfs2_reserve_new_inode(osb, &inode_ac);
 302         if (status < 0) {
 303                 if (status != -ENOSPC)
 304                         mlog_errno(status);
 305                 goto leave;
 306         }
 307
 308         inode = ocfs2_get_init_inode(dir, mode);
 309         if (IS_ERR(inode)) {
 310                 status = PTR_ERR(inode);
 311                 inode = NULL;
 312                 mlog_errno(status);
 313                 goto leave;
 314         }
 315
 316         /* get security xattr */
 317         status = ocfs2_init_security_get(inode, dir, &dentry->d_name, &si);
 318         if (status) {
 319                 if (status == -EOPNOTSUPP)
 320                         si.enable = 0;
 321                 else {
 322                         mlog_errno(status);
 323                         goto leave;
 324                 }
 325         }
 326
 327         /* calculate meta data/clusters for setting security and acl xattr */
 328         status = ocfs2_calc_xattr_init(dir, parent_fe_bh, mode,
 329                                        &si, &want_clusters,
 330                                        &xattr_credits, &want_meta);
 331         if (status < 0) {
 332                 mlog_errno(status);
 333                 goto leave;
 334         }
 335
 336         /* Reserve a cluster if creating an extent based directory. */
 337         if (S_ISDIR(mode) && !ocfs2_supports_inline_data(osb)) {
 338                 want_clusters += 1;
 339
 340                 /* Dir indexing requires extra space as well */
 341                 if (ocfs2_supports_indexed_dirs(osb))
 342                         want_meta++;
 343         }
 344
 345         status = ocfs2_reserve_new_metadata_blocks(osb, want_meta, &meta_ac);
 346         if (status < 0) {
 347                 if (status != -ENOSPC)
 348                         mlog_errno(status);
 349                 goto leave;
 350         }
 351
 352         status = ocfs2_reserve_clusters(osb, want_clusters, &data_ac);
 353         if (status < 0) {
 354                 if (status != -ENOSPC)
 355                         mlog_errno(status);
 356                 goto leave;
 357         }
 358
 359         handle = ocfs2_start_trans(osb, ocfs2_mknod_credits(osb->sb,
 360                                                             S_ISDIR(mode),
 361                                                             xattr_credits));
 362         if (IS_ERR(handle)) {
 363                 status = PTR_ERR(handle);
 364                 handle = NULL;
 365                 mlog_errno(status);
 366                 goto leave;
 367         }
 368
 369         /* Starting to change things, restart is no longer possible. */
 370         ocfs2_block_signals(&oldset);
 371         did_block_signals = 1;
 372
 373         status = dquot_alloc_inode(inode);
 374         if (status)
 375                 goto leave;
 376         did_quota_inode = 1;
 377
 378         /* do the real work now. */
 379         status = ocfs2_mknod_locked(osb, dir, inode, dev,
 380                                     &new_fe_bh, parent_fe_bh, handle,
 381                                     inode_ac);
 382         if (status < 0) {
 383                 mlog_errno(status);
 384                 goto leave;
 385         }
 386
 387         fe = (struct ocfs2_dinode *) new_fe_bh->b_data;
 388         if (S_ISDIR(mode)) {
 389                 status = ocfs2_fill_new_dir(osb, handle, dir, inode,
 390                                             new_fe_bh, data_ac, meta_ac);
 391                 if (status < 0) {
 392                         mlog_errno(status);
 393                         goto leave;
 394                 }
 395
 396                 status = ocfs2_journal_access_di(handle, INODE_CACHE(dir),
 397                                                  parent_fe_bh,
 398                                                  OCFS2_JOURNAL_ACCESS_WRITE);
 399                 if (status < 0) {
 400                         mlog_errno(status);
 401                         goto leave;
 402                 }
 403                 ocfs2_add_links_count(dirfe, 1);
 404                 ocfs2_journal_dirty(handle, parent_fe_bh);
 405                 inc_nlink(dir);
 406         }
 407
 408         status = ocfs2_init_acl(handle, inode, dir, new_fe_bh, parent_fe_bh,
 409                          meta_ac, data_ac);
 410
 411         if (status < 0) {
 412                 mlog_errno(status);
 413                 goto roll_back;
 414         }
 415
 416         if (si.enable) {
 417                 status = ocfs2_init_security_set(handle, inode, new_fe_bh, &si,
 418                                                  meta_ac, data_ac);
 419                 if (status < 0) {
 420                         mlog_errno(status);
 421                         goto roll_back;
 422                 }
 423         }
 424
 425         /*
 426          * Do this before adding the entry to the directory. We add
 427          * also set d_op after success so that ->d_iput() will cleanup
 428          * the dentry lock even if ocfs2_add_entry() fails below.
 429          */
 430         status = ocfs2_dentry_attach_lock(dentry, inode,
 431                                           OCFS2_I(dir)->ip_blkno);
 432         if (status) {
 433                 mlog_errno(status);
 434                 goto roll_back;
 435         }
 436
 437         dl = dentry->d_fsdata;
 438
 439         status = ocfs2_add_entry(handle, dentry, inode,
 440                                  OCFS2_I(inode)->ip_blkno, parent_fe_bh,
 441                                  &lookup);
 442         if (status < 0) {
 443                 mlog_errno(status);
 444                 goto roll_back;
 445         }
 446
 447         insert_inode_hash(inode);
 448         d_instantiate(dentry, inode);
 449         status = 0;
 450
 451 roll_back:
 452         if (status < 0 && S_ISDIR(mode)) {
 453                 ocfs2_add_links_count(dirfe, -1);
 454                 drop_nlink(dir);
 455         }
 456
 457 leave:
 458         if (status < 0 && did_quota_inode)
 459                 dquot_free_inode(inode);
 460         if (handle) {
 461                 if (status < 0 && fe)
 462                         ocfs2_set_links_count(fe, 0);
 463                 ocfs2_commit_trans(osb, handle);
 464         }
 465
 466         ocfs2_inode_unlock(dir, 1);
 467         if (did_block_signals)
 468                 ocfs2_unblock_signals(&oldset);
 469
 470         brelse(new_fe_bh);
 471         brelse(parent_fe_bh);
 472         kfree(si.value);
 473
 474         ocfs2_free_dir_lookup_result(&lookup);
 475
 476         if (inode_ac)
 477                 ocfs2_free_alloc_context(inode_ac);
 478
 479         if (data_ac)
 480                 ocfs2_free_alloc_context(data_ac);
 481
 482         if (meta_ac)
 483                 ocfs2_free_alloc_context(meta_ac);
 484
 485         /*
 486          * We should call iput after the i_mutex of the bitmap been
 487          * unlocked in ocfs2_free_alloc_context, or the
 488          * ocfs2_delete_inode will mutex_lock again.
 489          */
 490         if ((status < 0) && inode) {
 491                 if (dl)
 492                         ocfs2_cleanup_add_entry_failure(osb, dentry, inode);
 493
 494                 OCFS2_I(inode)->ip_flags |= OCFS2_INODE_SKIP_ORPHAN_DIR;
 495                 clear_nlink(inode);
 496                 iput(inode);
 497         }
 498
 499         if (status)
 500                 mlog_errno(status);
 501
 502         return status;
 503 }
 504
 505 static int __ocfs2_mknod_locked(struct inode *dir,
 506                                 struct inode *inode,
 507                                 dev_t dev,
 508                                 struct buffer_head **new_fe_bh,
 509                                 struct buffer_head *parent_fe_bh,
 510                                 handle_t *handle,
 511                                 struct ocfs2_alloc_context *inode_ac,
 512                                 u64 fe_blkno, u64 suballoc_loc, u16 suballoc_bit)
 513 {
 514         int status = 0;
 515         struct ocfs2_super *osb = OCFS2_SB(dir->i_sb);
 516         struct ocfs2_dinode *fe = NULL;
 517         struct ocfs2_extent_list *fel;
 518         u16 feat;
 519         struct ocfs2_inode_info *oi = OCFS2_I(inode);
 520         struct timespec64 ts;
 521
 522         *new_fe_bh = NULL;
 523
 524         /* populate as many fields early on as possible - many of
 525          * these are used by the support functions here and in
 526          * callers. */
 527         inode->i_ino = ino_from_blkno(osb->sb, fe_blkno);
 528         oi->ip_blkno = fe_blkno;
 529         spin_lock(&osb->osb_lock);
 530         inode->i_generation = osb->s_next_generation++;
 531         spin_unlock(&osb->osb_lock);
 532
 533         *new_fe_bh = sb_getblk(osb->sb, fe_blkno);
 534         if (!*new_fe_bh) {
 535                 status = -ENOMEM;
 536                 mlog_errno(status);
 537                 goto leave;
 538         }
 539         ocfs2_set_new_buffer_uptodate(INODE_CACHE(inode), *new_fe_bh);
 540
 541         status = ocfs2_journal_access_di(handle, INODE_CACHE(inode),
 542                                          *new_fe_bh,
 543                                          OCFS2_JOURNAL_ACCESS_CREATE);
 544         if (status < 0) {
 545                 mlog_errno(status);
 546                 goto leave;
 547         }
 548
 549         fe = (struct ocfs2_dinode *) (*new_fe_bh)->b_data;
 550         memset(fe, 0, osb->sb->s_blocksize);
 551
 552         fe->i_generation = cpu_to_le32(inode->i_generation);
 553         fe->i_fs_generation = cpu_to_le32(osb->fs_generation);
 554         fe->i_blkno = cpu_to_le64(fe_blkno);
 555         fe->i_suballoc_loc = cpu_to_le64(suballoc_loc);
 556         fe->i_suballoc_bit = cpu_to_le16(suballoc_bit);
 557         fe->i_suballoc_slot = cpu_to_le16(inode_ac->ac_alloc_slot);
 558         fe->i_uid = cpu_to_le32(i_uid_read(inode));
 559         fe->i_gid = cpu_to_le32(i_gid_read(inode));
 560         fe->i_mode = cpu_to_le16(inode->i_mode);
 561         if (S_ISCHR(inode->i_mode) || S_ISBLK(inode->i_mode))
 562                 fe->id1.dev1.i_rdev = cpu_to_le64(huge_encode_dev(dev));
 563
 564         ocfs2_set_links_count(fe, inode->i_nlink);
 565
 566         fe->i_last_eb_blk = 0;
 567         strcpy(fe->i_signature, OCFS2_INODE_SIGNATURE);
 568         fe->i_flags |= cpu_to_le32(OCFS2_VALID_FL);
 569         ktime_get_real_ts64(&ts);
 570         fe->i_atime = fe->i_ctime = fe->i_mtime =
 571                 cpu_to_le64(ts.tv_sec);
 572         fe->i_mtime_nsec = fe->i_ctime_nsec = fe->i_atime_nsec =
 573                 cpu_to_le32(ts.tv_nsec);
 574         fe->i_dtime = 0;
 575
 576         /*
 577          * If supported, directories start with inline data. If inline
 578          * isn't supported, but indexing is, we start them as indexed.
 579          */
 580         feat = le16_to_cpu(fe->i_dyn_features);
 581         if (S_ISDIR(inode->i_mode) && ocfs2_supports_inline_data(osb)) {
 582                 fe->i_dyn_features = cpu_to_le16(feat | OCFS2_INLINE_DATA_FL);
 583
 584                 fe->id2.i_data.id_count = cpu_to_le16(
 585                                 ocfs2_max_inline_data_with_xattr(osb->sb, fe));
 586         } else {
 587                 fel = &fe->id2.i_list;
 588                 fel->l_tree_depth = 0;
 589                 fel->l_next_free_rec = 0;
 590                 fel->l_count = cpu_to_le16(ocfs2_extent_recs_per_inode(osb->sb));
 591         }
 592
 593         ocfs2_journal_dirty(handle, *new_fe_bh);
 594
 595         ocfs2_populate_inode(inode, fe, 1);
 596         ocfs2_ci_set_new(osb, INODE_CACHE(inode));
 597         if (!ocfs2_mount_local(osb)) {
 598                 status = ocfs2_create_new_inode_locks(inode);
 599                 if (status < 0)
 600                         mlog_errno(status);
 601         }
 602
 603         ocfs2_update_inode_fsync_trans(handle, inode, 1);
 604
 605 leave:
 606         if (status < 0) {
 607                 if (*new_fe_bh) {
 608                         brelse(*new_fe_bh);
 609                         *new_fe_bh = NULL;
 610                 }
 611         }
 612
 613         if (status)
 614                 mlog_errno(status);
 615         return status;
 616 }
 617
 618 static int ocfs2_mknod_locked(struct ocfs2_super *osb,
 619                               struct inode *dir,
 620                               struct inode *inode,
 621                               dev_t dev,
 622                               struct buffer_head **new_fe_bh,
 623                               struct buffer_head *parent_fe_bh,
 624                               handle_t *handle,
 625                               struct ocfs2_alloc_context *inode_ac)
 626 {
 627         int status = 0;
 628         u64 suballoc_loc, fe_blkno = 0;
 629         u16 suballoc_bit;
 630
 631         *new_fe_bh = NULL;
 632
 633         status = ocfs2_claim_new_inode(handle, dir, parent_fe_bh,
 634                                        inode_ac, &suballoc_loc,
 635                                        &suballoc_bit, &fe_blkno);
 636         if (status < 0) {
 637                 mlog_errno(status);
 638                 return status;
 639         }
 640
 641         return __ocfs2_mknod_locked(dir, inode, dev, new_fe_bh,
 642                                     parent_fe_bh, handle, inode_ac,
 643                                     fe_blkno, suballoc_loc, suballoc_bit);
 644 }
 645
 646 static int ocfs2_mkdir(struct inode *dir,
 647                        struct dentry *dentry,
 648                        umode_t mode)
 649 {
 650         int ret;
 651
 652         trace_ocfs2_mkdir(dir, dentry, dentry->d_name.len, dentry->d_name.name,
 653                           OCFS2_I(dir)->ip_blkno, mode);
 654         ret = ocfs2_mknod(dir, dentry, mode | S_IFDIR, 0);
 655         if (ret)
 656                 mlog_errno(ret);
 657
 658         return ret;
 659 }
 660
 661 static int ocfs2_create(struct inode *dir,
 662                         struct dentry *dentry,
 663                         umode_t mode,
 664                         bool excl)
 665 {
 666         int ret;
 667
 668         trace_ocfs2_create(dir, dentry, dentry->d_name.len, dentry->d_name.name,
 669                            (unsigned long long)OCFS2_I(dir)->ip_blkno, mode);
 670         ret = ocfs2_mknod(dir, dentry, mode | S_IFREG, 0);
 671         if (ret)
 672                 mlog_errno(ret);
 673
 674         return ret;
 675 }
 676
 677 static int ocfs2_link(struct dentry *old_dentry,
 678                       struct inode *dir,
 679                       struct dentry *dentry)
 680 {
 681         handle_t *handle;
 682         struct inode *inode = d_inode(old_dentry);
 683         struct inode *old_dir = d_inode(old_dentry->d_parent);
 684         int err;
 685         struct buffer_head *fe_bh = NULL;
 686         struct buffer_head *old_dir_bh = NULL;
 687         struct buffer_head *parent_fe_bh = NULL;
 688         struct ocfs2_dinode *fe = NULL;
 689         struct ocfs2_super *osb = OCFS2_SB(dir->i_sb);
 690         struct ocfs2_dir_lookup_result lookup = { NULL, };
 691         sigset_t oldset;
 692         u64 old_de_ino;
 693
 694         trace_ocfs2_link((unsigned long long)OCFS2_I(inode)->ip_blkno,
 695                          old_dentry->d_name.len, old_dentry->d_name.name,
 696                          dentry->d_name.len, dentry->d_name.name);
 697
 698         if (S_ISDIR(inode->i_mode))
 699                 return -EPERM;
 700
 701         err = dquot_initialize(dir);
 702         if (err) {
 703                 mlog_errno(err);
 704                 return err;
 705         }
 706
 707         err = ocfs2_double_lock(osb, &old_dir_bh, old_dir,
 708                         &parent_fe_bh, dir, 0);
 709         if (err < 0) {
 710                 if (err != -ENOENT)
 711                         mlog_errno(err);
 712                 return err;
 713         }
 714
 715         /* make sure both dirs have bhs
 716          * get an extra ref on old_dir_bh if old==new */
 717         if (!parent_fe_bh) {
 718                 if (old_dir_bh) {
 719                         parent_fe_bh = old_dir_bh;
 720                         get_bh(parent_fe_bh);
 721                 } else {
 722                         mlog(ML_ERROR, "%s: no old_dir_bh!\n", osb->uuid_str);
 723                         err = -EIO;
 724                         goto out;
 725                 }
 726         }
 727
 728         if (!dir->i_nlink) {
 729                 err = -ENOENT;
 730                 goto out;
 731         }
 732
 733         err = ocfs2_lookup_ino_from_name(old_dir, old_dentry->d_name.name,
 734                         old_dentry->d_name.len, &old_de_ino);
 735         if (err) {
 736                 err = -ENOENT;
 737                 goto out;
 738         }
 739
 740         /*
 741          * Check whether another node removed the source inode while we
 742          * were in the vfs.
 743          */
 744         if (old_de_ino != OCFS2_I(inode)->ip_blkno) {
 745                 err = -ENOENT;
 746                 goto out;
 747         }
 748
 749         err = ocfs2_check_dir_for_entry(dir, dentry->d_name.name,
 750                                         dentry->d_name.len);
 751         if (err)
 752                 goto out;
 753
 754         err = ocfs2_prepare_dir_for_insert(osb, dir, parent_fe_bh,
 755                                            dentry->d_name.name,
 756                                            dentry->d_name.len, &lookup);
 757         if (err < 0) {
 758                 mlog_errno(err);
 759                 goto out;
 760         }
 761
 762         err = ocfs2_inode_lock(inode, &fe_bh, 1);
 763         if (err < 0) {
 764                 if (err != -ENOENT)
 765                         mlog_errno(err);
 766                 goto out;
 767         }
 768
 769         fe = (struct ocfs2_dinode *) fe_bh->b_data;
 770         if (ocfs2_read_links_count(fe) >= ocfs2_link_max(osb)) {
 771                 err = -EMLINK;
 772                 goto out_unlock_inode;
 773         }
 774
 775         handle = ocfs2_start_trans(osb, ocfs2_link_credits(osb->sb));
 776         if (IS_ERR(handle)) {
 777                 err = PTR_ERR(handle);
 778                 handle = NULL;
 779                 mlog_errno(err);
 780                 goto out_unlock_inode;
 781         }
 782
 783         /* Starting to change things, restart is no longer possible. */
 784         ocfs2_block_signals(&oldset);
 785
 786         err = ocfs2_journal_access_di(handle, INODE_CACHE(inode), fe_bh,
 787                                       OCFS2_JOURNAL_ACCESS_WRITE);
 788         if (err < 0) {
 789                 mlog_errno(err);
 790                 goto out_commit;
 791         }
 792
 793         inc_nlink(inode);
 794         inode->i_ctime = current_time(inode);
 795         ocfs2_set_links_count(fe, inode->i_nlink);
 796         fe->i_ctime = cpu_to_le64(inode->i_ctime.tv_sec);
 797         fe->i_ctime_nsec = cpu_to_le32(inode->i_ctime.tv_nsec);
 798         ocfs2_journal_dirty(handle, fe_bh);
 799
 800         err = ocfs2_add_entry(handle, dentry, inode,
 801                               OCFS2_I(inode)->ip_blkno,
 802                               parent_fe_bh, &lookup);
 803         if (err) {
 804                 ocfs2_add_links_count(fe, -1);
 805                 drop_nlink(inode);
 806                 mlog_errno(err);
 807                 goto out_commit;
 808         }
 809
 810         err = ocfs2_dentry_attach_lock(dentry, inode, OCFS2_I(dir)->ip_blkno);
 811         if (err) {
 812                 mlog_errno(err);
 813                 goto out_commit;
 814         }
 815
 816         ihold(inode);
 817         d_instantiate(dentry, inode);
 818
 819 out_commit:
 820         ocfs2_commit_trans(osb, handle);
 821         ocfs2_unblock_signals(&oldset);
 822 out_unlock_inode:
 823         ocfs2_inode_unlock(inode, 1);
 824
 825 out:
 826         ocfs2_double_unlock(old_dir, dir);
 827
 828         brelse(fe_bh);
 829         brelse(parent_fe_bh);
 830         brelse(old_dir_bh);
 831
 832         ocfs2_free_dir_lookup_result(&lookup);
 833
 834         if (err)
 835                 mlog_errno(err);
 836
 837         return err;
 838 }
 839
 840 /*
 841  * Takes and drops an exclusive lock on the given dentry. This will
 842  * force other nodes to drop it.
 843  */
 844 static int ocfs2_remote_dentry_delete(struct dentry *dentry)
 845 {
 846         int ret;
 847
 848         ret = ocfs2_dentry_lock(dentry, 1);
 849         if (ret)
 850                 mlog_errno(ret);
 851         else
 852                 ocfs2_dentry_unlock(dentry, 1);
 853
 854         return ret;
 855 }
 856
 857 static inline int ocfs2_inode_is_unlinkable(struct inode *inode)
 858 {
 859         if (S_ISDIR(inode->i_mode)) {
 860                 if (inode->i_nlink == 2)
 861                         return 1;
 862                 return 0;
 863         }
 864
 865         if (inode->i_nlink == 1)
 866                 return 1;
 867         return 0;
 868 }
 869
 870 static int ocfs2_unlink(struct inode *dir,
 871                         struct dentry *dentry)
 872 {
 873         int status;
 874         int child_locked = 0;
 875         bool is_unlinkable = false;
 876         struct inode *inode = d_inode(dentry);
 877         struct inode *orphan_dir = NULL;
 878         struct ocfs2_super *osb = OCFS2_SB(dir->i_sb);
 879         u64 blkno;
 880         struct ocfs2_dinode *fe = NULL;
 881         struct buffer_head *fe_bh = NULL;
 882         struct buffer_head *parent_node_bh = NULL;
 883         handle_t *handle = NULL;
 884         char orphan_name[OCFS2_ORPHAN_NAMELEN + 1];
 885         struct ocfs2_dir_lookup_result lookup = { NULL, };
 886         struct ocfs2_dir_lookup_result orphan_insert = { NULL, };
 887
 888         trace_ocfs2_unlink(dir, dentry, dentry->d_name.len,
 889                            dentry->d_name.name,
 890                            (unsigned long long)OCFS2_I(dir)->ip_blkno,
 891                            (unsigned long long)OCFS2_I(inode)->ip_blkno);
 892
 893         status = dquot_initialize(dir);
 894         if (status) {
 895                 mlog_errno(status);
 896                 return status;
 897         }
 898
 899         BUG_ON(d_inode(dentry->d_parent) != dir);
 900
 901         if (inode == osb->root_inode)
 902                 return -EPERM;
 903
 904         status = ocfs2_inode_lock_nested(dir, &parent_node_bh, 1,
 905                                          OI_LS_PARENT);
 906         if (status < 0) {
 907                 if (status != -ENOENT)
 908                         mlog_errno(status);
 909                 return status;
 910         }
 911
 912         status = ocfs2_find_files_on_disk(dentry->d_name.name,
 913                                           dentry->d_name.len, &blkno, dir,
 914                                           &lookup);
 915         if (status < 0) {
 916                 if (status != -ENOENT)
 917                         mlog_errno(status);
 918                 goto leave;
 919         }
 920
 921         if (OCFS2_I(inode)->ip_blkno != blkno) {
 922                 status = -ENOENT;
 923
 924                 trace_ocfs2_unlink_noent(
 925                                 (unsigned long long)OCFS2_I(inode)->ip_blkno,
 926                                 (unsigned long long)blkno,
 927                                 OCFS2_I(inode)->ip_flags);
 928                 goto leave;
 929         }
 930
 931         status = ocfs2_inode_lock(inode, &fe_bh, 1);
 932         if (status < 0) {
 933                 if (status != -ENOENT)
 934                         mlog_errno(status);
 935                 goto leave;
 936         }
 937         child_locked = 1;
 938
 939         if (S_ISDIR(inode->i_mode)) {
 940                 if (inode->i_nlink != 2 || !ocfs2_empty_dir(inode)) {
 941                         status = -ENOTEMPTY;
 942                         goto leave;
 943                 }
 944         }
 945
 946         status = ocfs2_remote_dentry_delete(dentry);
 947         if (status < 0) {
 948                 /* This remote delete should succeed under all normal
 949                  * circumstances. */
 950                 mlog_errno(status);
 951                 goto leave;
 952         }
 953
 954         if (ocfs2_inode_is_unlinkable(inode)) {
 955                 status = ocfs2_prepare_orphan_dir(osb, &orphan_dir,
 956                                                   OCFS2_I(inode)->ip_blkno,
 957                                                   orphan_name, &orphan_insert,
 958                                                   false);
 959                 if (status < 0) {
 960                         mlog_errno(status);
 961                         goto leave;
 962                 }
 963                 is_unlinkable = true;
 964         }
 965
 966         handle = ocfs2_start_trans(osb, ocfs2_unlink_credits(osb->sb));
 967         if (IS_ERR(handle)) {
 968                 status = PTR_ERR(handle);
 969                 handle = NULL;
 970                 mlog_errno(status);
 971                 goto leave;
 972         }
 973
 974         status = ocfs2_journal_access_di(handle, INODE_CACHE(inode), fe_bh,
 975                                          OCFS2_JOURNAL_ACCESS_WRITE);
 976         if (status < 0) {
 977                 mlog_errno(status);
 978                 goto leave;
 979         }
 980
 981         fe = (struct ocfs2_dinode *) fe_bh->b_data;
 982
 983         /* delete the name from the parent dir */
 984         status = ocfs2_delete_entry(handle, dir, &lookup);
 985         if (status < 0) {
 986                 mlog_errno(status);
 987                 goto leave;
 988         }
 989
 990         if (S_ISDIR(inode->i_mode))
 991                 drop_nlink(inode);
 992         drop_nlink(inode);
 993         ocfs2_set_links_count(fe, inode->i_nlink);
 994         ocfs2_journal_dirty(handle, fe_bh);
 995
 996         dir->i_ctime = dir->i_mtime = current_time(dir);
 997         if (S_ISDIR(inode->i_mode))
 998                 drop_nlink(dir);
 999
1000         status = ocfs2_mark_inode_dirty(handle, dir, parent_node_bh);
1001         if (status < 0) {
1002                 mlog_errno(status);
1003                 if (S_ISDIR(inode->i_mode))
1004                         inc_nlink(dir);
1005                 goto leave;
1006         }
1007
1008         if (is_unlinkable) {
1009                 status = ocfs2_orphan_add(osb, handle, inode, fe_bh,
1010                                 orphan_name, &orphan_insert, orphan_dir, false);
1011                 if (status < 0)
1012                         mlog_errno(status);
1013         }
1014
1015 leave:
1016         if (handle)
1017                 ocfs2_commit_trans(osb, handle);
1018
1019         if (orphan_dir) {
1020                 /* This was locked for us in ocfs2_prepare_orphan_dir() */
1021                 ocfs2_inode_unlock(orphan_dir, 1);
1022                 inode_unlock(orphan_dir);
1023                 iput(orphan_dir);
1024         }
1025
1026         if (child_locked)
1027                 ocfs2_inode_unlock(inode, 1);
1028
1029         ocfs2_inode_unlock(dir, 1);
1030
1031         brelse(fe_bh);
1032         brelse(parent_node_bh);
1033
1034         ocfs2_free_dir_lookup_result(&orphan_insert);
1035         ocfs2_free_dir_lookup_result(&lookup);
1036
1037         if (status && (status != -ENOTEMPTY) && (status != -ENOENT))
1038                 mlog_errno(status);
1039
1040         return status;
1041 }
1042
1043 static int ocfs2_check_if_ancestor(struct ocfs2_super *osb,
1044                 u64 src_inode_no, u64 dest_inode_no)
1045 {
1046         int ret = 0, i = 0;
1047         u64 parent_inode_no = 0;
1048         u64 child_inode_no = src_inode_no;
1049         struct inode *child_inode;
1050
1051 #define MAX_LOOKUP_TIMES 32
1052         while (1) {
1053                 child_inode = ocfs2_iget(osb, child_inode_no, 0, 0);
1054                 if (IS_ERR(child_inode)) {
1055                         ret = PTR_ERR(child_inode);
1056                         break;
1057                 }
1058
1059                 ret = ocfs2_inode_lock(child_inode, NULL, 0);
1060                 if (ret < 0) {
1061                         iput(child_inode);
1062                         if (ret != -ENOENT)
1063                                 mlog_errno(ret);
1064                         break;
1065                 }
1066
1067                 ret = ocfs2_lookup_ino_from_name(child_inode, "..", 2,
1068                                 &parent_inode_no);
1069                 ocfs2_inode_unlock(child_inode, 0);
1070                 iput(child_inode);
1071                 if (ret < 0) {
1072                         ret = -ENOENT;
1073                         break;
1074                 }
1075
1076                 if (parent_inode_no == dest_inode_no) {
1077                         ret = 1;
1078                         break;
1079                 }
1080
1081                 if (parent_inode_no == osb->root_inode->i_ino) {
1082                         ret = 0;
1083                         break;
1084                 }
1085
1086                 child_inode_no = parent_inode_no;
1087
1088                 if (++i >= MAX_LOOKUP_TIMES) {
1089                         mlog(ML_NOTICE, "max lookup times reached, filesystem "
1090                                         "may have nested directories, "
1091                                         "src inode: %llu, dest inode: %llu.\n",
1092                                         (unsigned long long)src_inode_no,
1093                                         (unsigned long long)dest_inode_no);
1094                         ret = 0;
1095                         break;
1096                 }
1097         }
1098
1099         return ret;
1100 }
1101
1102 /*
1103  * The only place this should be used is rename and link!
1104  * if they have the same id, then the 1st one is the only one locked.
1105  */
1106 static int ocfs2_double_lock(struct ocfs2_super *osb,
1107                              struct buffer_head **bh1,
1108                              struct inode *inode1,
1109                              struct buffer_head **bh2,
1110                              struct inode *inode2,
1111                              int rename)
1112 {
1113         int status;
1114         int inode1_is_ancestor, inode2_is_ancestor;
1115         struct ocfs2_inode_info *oi1 = OCFS2_I(inode1);
1116         struct ocfs2_inode_info *oi2 = OCFS2_I(inode2);
1117
1118         trace_ocfs2_double_lock((unsigned long long)oi1->ip_blkno,
1119                                 (unsigned long long)oi2->ip_blkno);
1120
1121         if (*bh1)
1122                 *bh1 = NULL;
1123         if (*bh2)
1124                 *bh2 = NULL;
1125
1126         /* we always want to lock the one with the lower lockid first.
1127          * and if they are nested, we lock ancestor first */
1128         if (oi1->ip_blkno != oi2->ip_blkno) {
1129                 inode1_is_ancestor = ocfs2_check_if_ancestor(osb, oi2->ip_blkno,
1130                                 oi1->ip_blkno);
1131                 if (inode1_is_ancestor < 0) {
1132                         status = inode1_is_ancestor;
1133                         goto bail;
1134                 }
1135
1136                 inode2_is_ancestor = ocfs2_check_if_ancestor(osb, oi1->ip_blkno,
1137                                 oi2->ip_blkno);
1138                 if (inode2_is_ancestor < 0) {
1139                         status = inode2_is_ancestor;
1140                         goto bail;
1141                 }
1142
1143                 if ((inode1_is_ancestor == 1) ||
1144                                 (oi1->ip_blkno < oi2->ip_blkno &&
1145                                 inode2_is_ancestor == 0)) {
1146                         /* switch id1 and id2 around */
1147                         swap(bh2, bh1);
1148                         swap(inode2, inode1);
1149                 }
1150                 /* lock id2 */
1151                 status = ocfs2_inode_lock_nested(inode2, bh2, 1,
1152                                 rename == 1 ? OI_LS_RENAME1 : OI_LS_PARENT);
1153                 if (status < 0) {
1154                         if (status != -ENOENT)
1155                                 mlog_errno(status);
1156                         goto bail;
1157                 }
1158         }
1159
1160         /* lock id1 */
1161         status = ocfs2_inode_lock_nested(inode1, bh1, 1,
1162                         rename == 1 ?  OI_LS_RENAME2 : OI_LS_PARENT);
1163         if (status < 0) {
1164                 /*
1165                  * An error return must mean that no cluster locks
1166                  * were held on function exit.
1167                  */
1168                 if (oi1->ip_blkno != oi2->ip_blkno) {
1169                         ocfs2_inode_unlock(inode2, 1);
1170                         brelse(*bh2);
1171                         *bh2 = NULL;
1172                 }
1173
1174                 if (status != -ENOENT)
1175                         mlog_errno(status);
1176         }
1177
1178         trace_ocfs2_double_lock_end(
1179                         (unsigned long long)oi1->ip_blkno,
1180                         (unsigned long long)oi2->ip_blkno);
1181
1182 bail:
1183         if (status)
1184                 mlog_errno(status);
1185         return status;
1186 }
1187
1188 static void ocfs2_double_unlock(struct inode *inode1, struct inode *inode2)
1189 {
1190         ocfs2_inode_unlock(inode1, 1);
1191
1192         if (inode1 != inode2)
1193                 ocfs2_inode_unlock(inode2, 1);
1194 }
1195
1196 static int ocfs2_rename(struct inode *old_dir,
1197                         struct dentry *old_dentry,
1198                         struct inode *new_dir,
1199                         struct dentry *new_dentry,
1200                         unsigned int flags)
1201 {
1202         int status = 0, rename_lock = 0, parents_locked = 0, target_exists = 0;
1203         int old_child_locked = 0, new_child_locked = 0, update_dot_dot = 0;
1204         struct inode *old_inode = d_inode(old_dentry);
1205         struct inode *new_inode = d_inode(new_dentry);
1206         struct inode *orphan_dir = NULL;
1207         struct ocfs2_dinode *newfe = NULL;
1208         char orphan_name[OCFS2_ORPHAN_NAMELEN + 1];
1209         struct buffer_head *newfe_bh = NULL;
1210         struct buffer_head *old_inode_bh = NULL;
1211         struct ocfs2_super *osb = NULL;
1212         u64 newfe_blkno, old_de_ino;
1213         handle_t *handle = NULL;
1214         struct buffer_head *old_dir_bh = NULL;
1215         struct buffer_head *new_dir_bh = NULL;
1216         u32 old_dir_nlink = old_dir->i_nlink;
1217         struct ocfs2_dinode *old_di;
1218         struct ocfs2_dir_lookup_result old_inode_dot_dot_res = { NULL, };
1219         struct ocfs2_dir_lookup_result target_lookup_res = { NULL, };
1220         struct ocfs2_dir_lookup_result old_entry_lookup = { NULL, };
1221         struct ocfs2_dir_lookup_result orphan_insert = { NULL, };
1222         struct ocfs2_dir_lookup_result target_insert = { NULL, };
1223         bool should_add_orphan = false;
1224
1225         if (flags)
1226                 return -EINVAL;
1227
1228         /* At some point it might be nice to break this function up a
1229          * bit. */
1230
1231         trace_ocfs2_rename(old_dir, old_dentry, new_dir, new_dentry,
1232                            old_dentry->d_name.len, old_dentry->d_name.name,
1233                            new_dentry->d_name.len, new_dentry->d_name.name);
1234
1235         status = dquot_initialize(old_dir);
1236         if (status) {
1237                 mlog_errno(status);
1238                 goto bail;
1239         }
1240         status = dquot_initialize(new_dir);
1241         if (status) {
1242                 mlog_errno(status);
1243                 goto bail;
1244         }
1245
1246         osb = OCFS2_SB(old_dir->i_sb);
1247
1248         if (new_inode) {
1249                 if (!igrab(new_inode))
1250                         BUG();
1251         }
1252
1253         /* Assume a directory hierarchy thusly:
1254          * a/b/c
1255          * a/d
1256          * a,b,c, and d are all directories.
1257          *
1258          * from cwd of 'a' on both nodes:
1259          * node1: mv b/c d
1260          * node2: mv d   b/c
1261          *
1262          * And that's why, just like the VFS, we need a file system
1263          * rename lock. */
1264         if (old_dir != new_dir && S_ISDIR(old_inode->i_mode)) {
1265                 status = ocfs2_rename_lock(osb);
1266                 if (status < 0) {
1267                         mlog_errno(status);
1268                         goto bail;
1269                 }
1270                 rename_lock = 1;
1271
1272                 /* here we cannot guarantee the inodes haven't just been
1273                  * changed, so check if they are nested again */
1274                 status = ocfs2_check_if_ancestor(osb, new_dir->i_ino,
1275                                 old_inode->i_ino);
1276                 if (status < 0) {
1277                         mlog_errno(status);
1278                         goto bail;
1279                 } else if (status == 1) {
1280                         status = -EPERM;
1281                         trace_ocfs2_rename_not_permitted(
1282                                         (unsigned long long)old_inode->i_ino,
1283                                         (unsigned long long)new_dir->i_ino);
1284                         goto bail;
1285                 }
1286         }
1287
1288         /* if old and new are the same, this'll just do one lock. */
1289         status = ocfs2_double_lock(osb, &old_dir_bh, old_dir,
1290                                    &new_dir_bh, new_dir, 1);
1291         if (status < 0) {
1292                 mlog_errno(status);
1293                 goto bail;
1294         }
1295         parents_locked = 1;
1296
1297         if (!new_dir->i_nlink) {
1298                 status = -EACCES;
1299                 goto bail;
1300         }
1301
1302         /* make sure both dirs have bhs
1303          * get an extra ref on old_dir_bh if old==new */
1304         if (!new_dir_bh) {
1305                 if (old_dir_bh) {
1306                         new_dir_bh = old_dir_bh;
1307                         get_bh(new_dir_bh);
1308                 } else {
1309                         mlog(ML_ERROR, "no old_dir_bh!\n");
1310                         status = -EIO;
1311                         goto bail;
1312                 }
1313         }
1314
1315         /*
1316          * Aside from allowing a meta data update, the locking here
1317          * also ensures that the downconvert thread on other nodes
1318          * won't have to concurrently downconvert the inode and the
1319          * dentry locks.
1320          */
1321         status = ocfs2_inode_lock_nested(old_inode, &old_inode_bh, 1,
1322                                          OI_LS_PARENT);
1323         if (status < 0) {
1324                 if (status != -ENOENT)
1325                         mlog_errno(status);
1326                 goto bail;
1327         }
1328         old_child_locked = 1;
1329
1330         status = ocfs2_remote_dentry_delete(old_dentry);
1331         if (status < 0) {
1332                 mlog_errno(status);
1333                 goto bail;
1334         }
1335
1336         if (S_ISDIR(old_inode->i_mode)) {
1337                 u64 old_inode_parent;
1338
1339                 update_dot_dot = 1;
1340                 status = ocfs2_find_files_on_disk("..", 2, &old_inode_parent,
1341                                                   old_inode,
1342                                                   &old_inode_dot_dot_res);
1343                 if (status) {
1344                         status = -EIO;
1345                         goto bail;
1346                 }
1347
1348                 if (old_inode_parent != OCFS2_I(old_dir)->ip_blkno) {
1349                         status = -EIO;
1350                         goto bail;
1351                 }
1352
1353                 if (!new_inode && new_dir != old_dir &&
1354                     new_dir->i_nlink >= ocfs2_link_max(osb)) {
1355                         status = -EMLINK;
1356                         goto bail;
1357                 }
1358         }
1359
1360         status = ocfs2_lookup_ino_from_name(old_dir, old_dentry->d_name.name,
1361                                             old_dentry->d_name.len,
1362                                             &old_de_ino);
1363         if (status) {
1364                 status = -ENOENT;
1365                 goto bail;
1366         }
1367
1368         /*
1369          *  Check for inode number is _not_ due to possible IO errors.
1370          *  We might rmdir the source, keep it as pwd of some process
1371          *  and merrily kill the link to whatever was created under the
1372          *  same name. Goodbye sticky bit ;-<
1373          */
1374         if (old_de_ino != OCFS2_I(old_inode)->ip_blkno) {
1375                 status = -ENOENT;
1376                 goto bail;
1377         }
1378
1379         /* check if the target already exists (in which case we need
1380          * to delete it */
1381         status = ocfs2_find_files_on_disk(new_dentry->d_name.name,
1382                                           new_dentry->d_name.len,
1383                                           &newfe_blkno, new_dir,
1384                                           &target_lookup_res);
1385         /* The only error we allow here is -ENOENT because the new
1386          * file not existing is perfectly valid. */
1387         if ((status < 0) && (status != -ENOENT)) {
1388                 /* If we cannot find the file specified we should just */
1389                 /* return the error... */
1390                 mlog_errno(status);
1391                 goto bail;
1392         }
1393         if (status == 0)
1394                 target_exists = 1;
1395
1396         if (!target_exists && new_inode) {
1397                 /*
1398                  * Target was unlinked by another node while we were
1399                  * waiting to get to ocfs2_rename(). There isn't
1400                  * anything we can do here to help the situation, so
1401                  * bubble up the appropriate error.
1402                  */
1403                 status = -ENOENT;
1404                 goto bail;
1405         }
1406
1407         /* In case we need to overwrite an existing file, we blow it
1408          * away first */
1409         if (target_exists) {
1410                 /* VFS didn't think there existed an inode here, but
1411                  * someone else in the cluster must have raced our
1412                  * rename to create one. Today we error cleanly, in
1413                  * the future we should consider calling iget to build
1414                  * a new struct inode for this entry. */
1415                 if (!new_inode) {
1416                         status = -EACCES;
1417
1418                         trace_ocfs2_rename_target_exists(new_dentry->d_name.len,
1419                                                 new_dentry->d_name.name);
1420                         goto bail;
1421                 }
1422
1423                 if (OCFS2_I(new_inode)->ip_blkno != newfe_blkno) {
1424                         status = -EACCES;
1425
1426                         trace_ocfs2_rename_disagree(
1427                              (unsigned long long)OCFS2_I(new_inode)->ip_blkno,
1428                              (unsigned long long)newfe_blkno,
1429                              OCFS2_I(new_inode)->ip_flags);
1430                         goto bail;
1431                 }
1432
1433                 status = ocfs2_inode_lock(new_inode, &newfe_bh, 1);
1434                 if (status < 0) {
1435                         if (status != -ENOENT)
1436                                 mlog_errno(status);
1437                         goto bail;
1438                 }
1439                 new_child_locked = 1;
1440
1441                 status = ocfs2_remote_dentry_delete(new_dentry);
1442                 if (status < 0) {
1443                         mlog_errno(status);
1444                         goto bail;
1445                 }
1446
1447                 newfe = (struct ocfs2_dinode *) newfe_bh->b_data;
1448
1449                 trace_ocfs2_rename_over_existing(
1450                      (unsigned long long)newfe_blkno, newfe_bh, newfe_bh ?
1451                      (unsigned long long)newfe_bh->b_blocknr : 0ULL);
1452
1453                 if (S_ISDIR(new_inode->i_mode) || (new_inode->i_nlink == 1)) {
1454                         status = ocfs2_prepare_orphan_dir(osb, &orphan_dir,
1455                                                 OCFS2_I(new_inode)->ip_blkno,
1456                                                 orphan_name, &orphan_insert,
1457                                                 false);
1458                         if (status < 0) {
1459                                 mlog_errno(status);
1460                                 goto bail;
1461                         }
1462                         should_add_orphan = true;
1463                 }
1464         } else {
1465                 BUG_ON(d_inode(new_dentry->d_parent) != new_dir);
1466
1467                 status = ocfs2_check_dir_for_entry(new_dir,
1468                                                    new_dentry->d_name.name,
1469                                                    new_dentry->d_name.len);
1470                 if (status)
1471                         goto bail;
1472
1473                 status = ocfs2_prepare_dir_for_insert(osb, new_dir, new_dir_bh,
1474                                                       new_dentry->d_name.name,
1475                                                       new_dentry->d_name.len,
1476                                                       &target_insert);
1477                 if (status < 0) {
1478                         mlog_errno(status);
1479                         goto bail;
1480                 }
1481         }
1482
1483         handle = ocfs2_start_trans(osb, ocfs2_rename_credits(osb->sb));
1484         if (IS_ERR(handle)) {
1485                 status = PTR_ERR(handle);
1486                 handle = NULL;
1487                 mlog_errno(status);
1488                 goto bail;
1489         }
1490
1491         if (target_exists) {
1492                 if (S_ISDIR(new_inode->i_mode)) {
1493                         if (new_inode->i_nlink != 2 ||
1494                             !ocfs2_empty_dir(new_inode)) {
1495                                 status = -ENOTEMPTY;
1496                                 goto bail;
1497                         }
1498                 }
1499                 status = ocfs2_journal_access_di(handle, INODE_CACHE(new_inode),
1500                                                  newfe_bh,
1501                                                  OCFS2_JOURNAL_ACCESS_WRITE);
1502                 if (status < 0) {
1503                         mlog_errno(status);
1504                         goto bail;
1505                 }
1506
1507                 /* change the dirent to point to the correct inode */
1508                 status = ocfs2_update_entry(new_dir, handle, &target_lookup_res,
1509                                             old_inode);
1510                 if (status < 0) {
1511                         mlog_errno(status);
1512                         goto bail;
1513                 }
1514                 inode_inc_iversion(new_dir);
1515
1516                 if (S_ISDIR(new_inode->i_mode))
1517                         ocfs2_set_links_count(newfe, 0);
1518                 else
1519                         ocfs2_add_links_count(newfe, -1);
1520                 ocfs2_journal_dirty(handle, newfe_bh);
1521                 if (should_add_orphan) {
1522                         status = ocfs2_orphan_add(osb, handle, new_inode,
1523                                         newfe_bh, orphan_name,
1524                                         &orphan_insert, orphan_dir, false);
1525                         if (status < 0) {
1526                                 mlog_errno(status);
1527                                 goto bail;
1528                         }
1529                 }
1530         } else {
1531                 /* if the name was not found in new_dir, add it now */
1532                 status = ocfs2_add_entry(handle, new_dentry, old_inode,
1533                                          OCFS2_I(old_inode)->ip_blkno,
1534                                          new_dir_bh, &target_insert);
1535                 if (status < 0) {
1536                         mlog_errno(status);
1537                         goto bail;
1538                 }
1539         }
1540
1541         old_inode->i_ctime = current_time(old_inode);
1542         mark_inode_dirty(old_inode);
1543
1544         status = ocfs2_journal_access_di(handle, INODE_CACHE(old_inode),
1545                                          old_inode_bh,
1546                                          OCFS2_JOURNAL_ACCESS_WRITE);
1547         if (status >= 0) {
1548                 old_di = (struct ocfs2_dinode *) old_inode_bh->b_data;
1549
1550                 old_di->i_ctime = cpu_to_le64(old_inode->i_ctime.tv_sec);
1551                 old_di->i_ctime_nsec = cpu_to_le32(old_inode->i_ctime.tv_nsec);
1552                 ocfs2_journal_dirty(handle, old_inode_bh);
1553         } else
1554                 mlog_errno(status);
1555
1556         /*
1557          * Now that the name has been added to new_dir, remove the old name.
1558          *
1559          * We don't keep any directory entry context around until now
1560          * because the insert might have changed the type of directory
1561          * we're dealing with.
1562          */
1563         status = ocfs2_find_entry(old_dentry->d_name.name,
1564                                   old_dentry->d_name.len, old_dir,
1565                                   &old_entry_lookup);
1566         if (status) {
1567                 if (!is_journal_aborted(osb->journal->j_journal)) {
1568                         ocfs2_error(osb->sb, "new entry %.*s is added, but old entry %.*s "
1569                                         "is not deleted.",
1570                                         new_dentry->d_name.len, new_dentry->d_name.name,
1571                                         old_dentry->d_name.len, old_dentry->d_name.name);
1572                 }
1573                 goto bail;
1574         }
1575
1576         status = ocfs2_delete_entry(handle, old_dir, &old_entry_lookup);
1577         if (status < 0) {
1578                 mlog_errno(status);
1579                 if (!is_journal_aborted(osb->journal->j_journal)) {
1580                         ocfs2_error(osb->sb, "new entry %.*s is added, but old entry %.*s "
1581                                         "is not deleted.",
1582                                         new_dentry->d_name.len, new_dentry->d_name.name,
1583                                         old_dentry->d_name.len, old_dentry->d_name.name);
1584                 }
1585                 goto bail;
1586         }
1587
1588         if (new_inode) {
1589                 drop_nlink(new_inode);
1590                 new_inode->i_ctime = current_time(new_inode);
1591         }
1592         old_dir->i_ctime = old_dir->i_mtime = current_time(old_dir);
1593
1594         if (update_dot_dot) {
1595                 status = ocfs2_update_entry(old_inode, handle,
1596                                             &old_inode_dot_dot_res, new_dir);
1597                 drop_nlink(old_dir);
1598                 if (new_inode) {
1599                         drop_nlink(new_inode);
1600                 } else {
1601                         inc_nlink(new_dir);
1602                         mark_inode_dirty(new_dir);
1603                 }
1604         }
1605         mark_inode_dirty(old_dir);
1606         ocfs2_mark_inode_dirty(handle, old_dir, old_dir_bh);
1607         if (new_inode) {
1608                 mark_inode_dirty(new_inode);
1609                 ocfs2_mark_inode_dirty(handle, new_inode, newfe_bh);
1610         }
1611
1612         if (old_dir != new_dir) {
1613                 /* Keep the same times on both directories.*/
1614                 new_dir->i_ctime = new_dir->i_mtime = old_dir->i_ctime;
1615
1616                 /*
1617                  * This will also pick up the i_nlink change from the
1618                  * block above.
1619                  */
1620                 ocfs2_mark_inode_dirty(handle, new_dir, new_dir_bh);
1621         }
1622
1623         if (old_dir_nlink != old_dir->i_nlink) {
1624                 if (!old_dir_bh) {
1625                         mlog(ML_ERROR, "need to change nlink for old dir "
1626                              "%llu from %d to %d but bh is NULL!\n",
1627                              (unsigned long long)OCFS2_I(old_dir)->ip_blkno,
1628                              (int)old_dir_nlink, old_dir->i_nlink);
1629                 } else {
1630                         struct ocfs2_dinode *fe;
1631                         status = ocfs2_journal_access_di(handle,
1632                                                          INODE_CACHE(old_dir),
1633                                                          old_dir_bh,
1634                                                          OCFS2_JOURNAL_ACCESS_WRITE);
1635                         fe = (struct ocfs2_dinode *) old_dir_bh->b_data;
1636                         ocfs2_set_links_count(fe, old_dir->i_nlink);
1637                         ocfs2_journal_dirty(handle, old_dir_bh);
1638                 }
1639         }
1640         ocfs2_dentry_move(old_dentry, new_dentry, old_dir, new_dir);
1641         status = 0;
1642 bail:
1643         if (handle)
1644                 ocfs2_commit_trans(osb, handle);
1645
1646         if (orphan_dir) {
1647                 /* This was locked for us in ocfs2_prepare_orphan_dir() */
1648                 ocfs2_inode_unlock(orphan_dir, 1);
1649                 inode_unlock(orphan_dir);
1650                 iput(orphan_dir);
1651         }
1652
1653         if (new_child_locked)
1654                 ocfs2_inode_unlock(new_inode, 1);
1655
1656         if (old_child_locked)
1657                 ocfs2_inode_unlock(old_inode, 1);
1658
1659         if (parents_locked)
1660                 ocfs2_double_unlock(old_dir, new_dir);
1661
1662         if (rename_lock)
1663                 ocfs2_rename_unlock(osb);
1664
1665         if (new_inode)
1666                 sync_mapping_buffers(old_inode->i_mapping);
1667
1668         iput(new_inode);
1669
1670         ocfs2_free_dir_lookup_result(&target_lookup_res);
1671         ocfs2_free_dir_lookup_result(&old_entry_lookup);
1672         ocfs2_free_dir_lookup_result(&old_inode_dot_dot_res);
1673         ocfs2_free_dir_lookup_result(&orphan_insert);
1674         ocfs2_free_dir_lookup_result(&target_insert);
1675
1676         brelse(newfe_bh);
1677         brelse(old_inode_bh);
1678         brelse(old_dir_bh);
1679         brelse(new_dir_bh);
1680
1681         if (status)
1682                 mlog_errno(status);
1683
1684         return status;
1685 }
1686
1687 /*
1688  * we expect i_size = strlen(symname). Copy symname into the file
1689  * data, including the null terminator.
1690  */
1691 static int ocfs2_create_symlink_data(struct ocfs2_super *osb,
1692                                      handle_t *handle,
1693                                      struct inode *inode,
1694                                      const char *symname)
1695 {
1696         struct buffer_head **bhs = NULL;
1697         const char *c;
1698         struct super_block *sb = osb->sb;
1699         u64 p_blkno, p_blocks;
1700         int virtual, blocks, status, i, bytes_left;
1701
1702         bytes_left = i_size_read(inode) + 1;
1703         /* we can't trust i_blocks because we're actually going to
1704          * write i_size + 1 bytes. */
1705         blocks = (bytes_left + sb->s_blocksize - 1) >> sb->s_blocksize_bits;
1706
1707         trace_ocfs2_create_symlink_data((unsigned long long)inode->i_blocks,
1708                                         i_size_read(inode), blocks);
1709
1710         /* Sanity check -- make sure we're going to fit. */
1711         if (bytes_left >
1712             ocfs2_clusters_to_bytes(sb, OCFS2_I(inode)->ip_clusters)) {
1713                 status = -EIO;
1714                 mlog_errno(status);
1715                 goto bail;
1716         }
1717
1718         bhs = kcalloc(blocks, sizeof(struct buffer_head *), GFP_KERNEL);
1719         if (!bhs) {
1720                 status = -ENOMEM;
1721                 mlog_errno(status);
1722                 goto bail;
1723         }
1724
1725         status = ocfs2_extent_map_get_blocks(inode, 0, &p_blkno, &p_blocks,
1726                                              NULL);
1727         if (status < 0) {
1728                 mlog_errno(status);
1729                 goto bail;
1730         }
1731
1732         /* links can never be larger than one cluster so we know this
1733          * is all going to be contiguous, but do a sanity check
1734          * anyway. */
1735         if ((p_blocks << sb->s_blocksize_bits) < bytes_left) {
1736                 status = -EIO;
1737                 mlog_errno(status);
1738                 goto bail;
1739         }
1740
1741         virtual = 0;
1742         while(bytes_left > 0) {
1743                 c = &symname[virtual * sb->s_blocksize];
1744
1745                 bhs[virtual] = sb_getblk(sb, p_blkno);
1746                 if (!bhs[virtual]) {
1747                         status = -ENOMEM;
1748                         mlog_errno(status);
1749                         goto bail;
1750                 }
1751                 ocfs2_set_new_buffer_uptodate(INODE_CACHE(inode),
1752                                               bhs[virtual]);
1753
1754                 status = ocfs2_journal_access(handle, INODE_CACHE(inode),
1755                                               bhs[virtual],
1756                                               OCFS2_JOURNAL_ACCESS_CREATE);
1757                 if (status < 0) {
1758                         mlog_errno(status);
1759                         goto bail;
1760                 }
1761
1762                 memset(bhs[virtual]->b_data, 0, sb->s_blocksize);
1763
1764                 memcpy(bhs[virtual]->b_data, c,
1765                        (bytes_left > sb->s_blocksize) ? sb->s_blocksize :
1766                        bytes_left);
1767
1768                 ocfs2_journal_dirty(handle, bhs[virtual]);
1769
1770                 virtual++;
1771                 p_blkno++;
1772                 bytes_left -= sb->s_blocksize;
1773         }
1774
1775         status = 0;
1776 bail:
1777
1778         if (bhs) {
1779                 for(i = 0; i < blocks; i++)
1780                         brelse(bhs[i]);
1781                 kfree(bhs);
1782         }
1783
1784         if (status)
1785                 mlog_errno(status);
1786         return status;
1787 }
1788
1789 static int ocfs2_symlink(struct inode *dir,
1790                          struct dentry *dentry,
1791                          const char *symname)
1792 {
1793         int status, l, credits;
1794         u64 newsize;
1795         struct ocfs2_super *osb = NULL;
1796         struct inode *inode = NULL;
1797         struct super_block *sb;
1798         struct buffer_head *new_fe_bh = NULL;
1799         struct buffer_head *parent_fe_bh = NULL;
1800         struct ocfs2_dinode *fe = NULL;
1801         struct ocfs2_dinode *dirfe;
1802         handle_t *handle = NULL;
1803         struct ocfs2_alloc_context *inode_ac = NULL;
1804         struct ocfs2_alloc_context *data_ac = NULL;
1805         struct ocfs2_alloc_context *xattr_ac = NULL;
1806         int want_clusters = 0;
1807         int xattr_credits = 0;
1808         struct ocfs2_security_xattr_info si = {
1809                 .name = NULL,
1810                 .enable = 1,
1811         };
1812         int did_quota = 0, did_quota_inode = 0;
1813         struct ocfs2_dir_lookup_result lookup = { NULL, };
1814         sigset_t oldset;
1815         int did_block_signals = 0;
1816         struct ocfs2_dentry_lock *dl = NULL;
1817
1818         trace_ocfs2_symlink_begin(dir, dentry, symname,
1819                                   dentry->d_name.len, dentry->d_name.name);
1820
1821         status = dquot_initialize(dir);
1822         if (status) {
1823                 mlog_errno(status);
1824                 goto bail;
1825         }
1826
1827         sb = dir->i_sb;
1828         osb = OCFS2_SB(sb);
1829
1830         l = strlen(symname) + 1;
1831
1832         credits = ocfs2_calc_symlink_credits(sb);
1833
1834         /* lock the parent directory */
1835         status = ocfs2_inode_lock(dir, &parent_fe_bh, 1);
1836         if (status < 0) {
1837                 if (status != -ENOENT)
1838                         mlog_errno(status);
1839                 return status;
1840         }
1841
1842         dirfe = (struct ocfs2_dinode *) parent_fe_bh->b_data;
1843         if (!ocfs2_read_links_count(dirfe)) {
1844                 /* can't make a file in a deleted directory. */
1845                 status = -ENOENT;
1846                 goto bail;
1847         }
1848
1849         status = ocfs2_check_dir_for_entry(dir, dentry->d_name.name,
1850                                            dentry->d_name.len);
1851         if (status)
1852                 goto bail;
1853
1854         status = ocfs2_prepare_dir_for_insert(osb, dir, parent_fe_bh,
1855                                               dentry->d_name.name,
1856                                               dentry->d_name.len, &lookup);
1857         if (status < 0) {
1858                 mlog_errno(status);
1859                 goto bail;
1860         }
1861
1862         status = ocfs2_reserve_new_inode(osb, &inode_ac);
1863         if (status < 0) {
1864                 if (status != -ENOSPC)
1865                         mlog_errno(status);
1866                 goto bail;
1867         }
1868
1869         inode = ocfs2_get_init_inode(dir, S_IFLNK | S_IRWXUGO);
1870         if (IS_ERR(inode)) {
1871                 status = PTR_ERR(inode);
1872                 inode = NULL;
1873                 mlog_errno(status);
1874                 goto bail;
1875         }
1876
1877         /* get security xattr */
1878         status = ocfs2_init_security_get(inode, dir, &dentry->d_name, &si);
1879         if (status) {
1880                 if (status == -EOPNOTSUPP)
1881                         si.enable = 0;
1882                 else {
1883                         mlog_errno(status);
1884                         goto bail;
1885                 }
1886         }
1887
1888         /* calculate meta data/clusters for setting security xattr */
1889         if (si.enable) {
1890                 status = ocfs2_calc_security_init(dir, &si, &want_clusters,
1891                                                   &xattr_credits, &xattr_ac);
1892                 if (status < 0) {
1893                         mlog_errno(status);
1894                         goto bail;
1895                 }
1896         }
1897
1898         /* don't reserve bitmap space for fast symlinks. */
1899         if (l > ocfs2_fast_symlink_chars(sb))
1900                 want_clusters += 1;
1901
1902         status = ocfs2_reserve_clusters(osb, want_clusters, &data_ac);
1903         if (status < 0) {
1904                 if (status != -ENOSPC)
1905                         mlog_errno(status);
1906                 goto bail;
1907         }
1908
1909         handle = ocfs2_start_trans(osb, credits + xattr_credits);
1910         if (IS_ERR(handle)) {
1911                 status = PTR_ERR(handle);
1912                 handle = NULL;
1913                 mlog_errno(status);
1914                 goto bail;
1915         }
1916
1917         /* Starting to change things, restart is no longer possible. */
1918         ocfs2_block_signals(&oldset);
1919         did_block_signals = 1;
1920
1921         status = dquot_alloc_inode(inode);
1922         if (status)
1923                 goto bail;
1924         did_quota_inode = 1;
1925
1926         trace_ocfs2_symlink_create(dir, dentry, dentry->d_name.len,
1927                                    dentry->d_name.name,
1928                                    (unsigned long long)OCFS2_I(dir)->ip_blkno,
1929                                    inode->i_mode);
1930
1931         status = ocfs2_mknod_locked(osb, dir, inode,
1932                                     0, &new_fe_bh, parent_fe_bh, handle,
1933                                     inode_ac);
1934         if (status < 0) {
1935                 mlog_errno(status);
1936                 goto bail;
1937         }
1938
1939         fe = (struct ocfs2_dinode *) new_fe_bh->b_data;
1940         inode->i_rdev = 0;
1941         newsize = l - 1;
1942         inode->i_op = &ocfs2_symlink_inode_operations;
1943         inode_nohighmem(inode);
1944         if (l > ocfs2_fast_symlink_chars(sb)) {
1945                 u32 offset = 0;
1946
1947                 status = dquot_alloc_space_nodirty(inode,
1948                     ocfs2_clusters_to_bytes(osb->sb, 1));
1949                 if (status)
1950                         goto bail;
1951                 did_quota = 1;
1952                 inode->i_mapping->a_ops = &ocfs2_aops;
1953                 status = ocfs2_add_inode_data(osb, inode, &offset, 1, 0,
1954                                               new_fe_bh,
1955                                               handle, data_ac, NULL,
1956                                               NULL);
1957                 if (status < 0) {
1958                         if (status != -ENOSPC && status != -EINTR) {
1959                                 mlog(ML_ERROR,
1960                                      "Failed to extend file to %llu\n",
1961                                      (unsigned long long)newsize);
1962                                 mlog_errno(status);
1963                                 status = -ENOSPC;
1964                         }
1965                         goto bail;
1966                 }
1967                 i_size_write(inode, newsize);
1968                 inode->i_blocks = ocfs2_inode_sector_count(inode);
1969         } else {
1970                 inode->i_mapping->a_ops = &ocfs2_fast_symlink_aops;
1971                 memcpy((char *) fe->id2.i_symlink, symname, l);
1972                 i_size_write(inode, newsize);
1973                 inode->i_blocks = 0;
1974         }
1975
1976         status = ocfs2_mark_inode_dirty(handle, inode, new_fe_bh);
1977         if (status < 0) {
1978                 mlog_errno(status);
1979                 goto bail;
1980         }
1981
1982         if (!ocfs2_inode_is_fast_symlink(inode)) {
1983                 status = ocfs2_create_symlink_data(osb, handle, inode,
1984                                                    symname);
1985                 if (status < 0) {
1986                         mlog_errno(status);
1987                         goto bail;
1988                 }
1989         }
1990
1991         if (si.enable) {
1992                 status = ocfs2_init_security_set(handle, inode, new_fe_bh, &si,
1993                                                  xattr_ac, data_ac);
1994                 if (status < 0) {
1995                         mlog_errno(status);
1996                         goto bail;
1997                 }
1998         }
1999
2000         /*
2001          * Do this before adding the entry to the directory. We add
2002          * also set d_op after success so that ->d_iput() will cleanup
2003          * the dentry lock even if ocfs2_add_entry() fails below.
2004          */
2005         status = ocfs2_dentry_attach_lock(dentry, inode, OCFS2_I(dir)->ip_blkno);
2006         if (status) {
2007                 mlog_errno(status);
2008                 goto bail;
2009         }
2010
2011         dl = dentry->d_fsdata;
2012
2013         status = ocfs2_add_entry(handle, dentry, inode,
2014                                  le64_to_cpu(fe->i_blkno), parent_fe_bh,
2015                                  &lookup);
2016         if (status < 0) {
2017                 mlog_errno(status);
2018                 goto bail;
2019         }
2020
2021         insert_inode_hash(inode);
2022         d_instantiate(dentry, inode);
2023 bail:
2024         if (status < 0 && did_quota)
2025                 dquot_free_space_nodirty(inode,
2026                                         ocfs2_clusters_to_bytes(osb->sb, 1));
2027         if (status < 0 && did_quota_inode)
2028                 dquot_free_inode(inode);
2029         if (handle) {
2030                 if (status < 0 && fe)
2031                         ocfs2_set_links_count(fe, 0);
2032                 ocfs2_commit_trans(osb, handle);
2033         }
2034
2035         ocfs2_inode_unlock(dir, 1);
2036         if (did_block_signals)
2037                 ocfs2_unblock_signals(&oldset);
2038
2039         brelse(new_fe_bh);
2040         brelse(parent_fe_bh);
2041         kfree(si.value);
2042         ocfs2_free_dir_lookup_result(&lookup);
2043         if (inode_ac)
2044                 ocfs2_free_alloc_context(inode_ac);
2045         if (data_ac)
2046                 ocfs2_free_alloc_context(data_ac);
2047         if (xattr_ac)
2048                 ocfs2_free_alloc_context(xattr_ac);
2049         if ((status < 0) && inode) {
2050                 if (dl)
2051                         ocfs2_cleanup_add_entry_failure(osb, dentry, inode);
2052
2053                 OCFS2_I(inode)->ip_flags |= OCFS2_INODE_SKIP_ORPHAN_DIR;
2054                 clear_nlink(inode);
2055                 iput(inode);
2056         }
2057
2058         if (status)
2059                 mlog_errno(status);
2060
2061         return status;
2062 }
2063
2064 static int ocfs2_blkno_stringify(u64 blkno, char *name)
2065 {
2066         int status, namelen;
2067
2068         namelen = snprintf(name, OCFS2_ORPHAN_NAMELEN + 1, "%016llx",
2069                            (long long)blkno);
2070         if (namelen <= 0) {
2071                 if (namelen)
2072                         status = namelen;
2073                 else
2074                         status = -EINVAL;
2075                 mlog_errno(status);
2076                 goto bail;
2077         }
2078         if (namelen != OCFS2_ORPHAN_NAMELEN) {
2079                 status = -EINVAL;
2080                 mlog_errno(status);
2081                 goto bail;
2082         }
2083
2084         trace_ocfs2_blkno_stringify(blkno, name, namelen);
2085
2086         status = 0;
2087 bail:
2088         if (status < 0)
2089                 mlog_errno(status);
2090         return status;
2091 }
2092
2093 static int ocfs2_lookup_lock_orphan_dir(struct ocfs2_super *osb,
2094                                         struct inode **ret_orphan_dir,
2095                                         struct buffer_head **ret_orphan_dir_bh)
2096 {
2097         struct inode *orphan_dir_inode;
2098         struct buffer_head *orphan_dir_bh = NULL;
2099         int ret = 0;
2100
2101         orphan_dir_inode = ocfs2_get_system_file_inode(osb,
2102                                                        ORPHAN_DIR_SYSTEM_INODE,
2103                                                        osb->slot_num);
2104         if (!orphan_dir_inode) {
2105                 ret = -ENOENT;
2106                 mlog_errno(ret);
2107                 return ret;
2108         }
2109
2110         inode_lock(orphan_dir_inode);
2111
2112         ret = ocfs2_inode_lock(orphan_dir_inode, &orphan_dir_bh, 1);
2113         if (ret < 0) {
2114                 inode_unlock(orphan_dir_inode);
2115                 iput(orphan_dir_inode);
2116
2117                 mlog_errno(ret);
2118                 return ret;
2119         }
2120
2121         *ret_orphan_dir = orphan_dir_inode;
2122         *ret_orphan_dir_bh = orphan_dir_bh;
2123
2124         return 0;
2125 }
2126
2127 static int __ocfs2_prepare_orphan_dir(struct inode *orphan_dir_inode,
2128                                       struct buffer_head *orphan_dir_bh,
2129                                       u64 blkno,
2130                                       char *name,
2131                                       struct ocfs2_dir_lookup_result *lookup,
2132                                       bool dio)
2133 {
2134         int ret;
2135         struct ocfs2_super *osb = OCFS2_SB(orphan_dir_inode->i_sb);
2136         int namelen = dio ?
2137                         (OCFS2_DIO_ORPHAN_PREFIX_LEN + OCFS2_ORPHAN_NAMELEN) :
2138                         OCFS2_ORPHAN_NAMELEN;
2139
2140         if (dio) {
2141                 ret = snprintf(name, OCFS2_DIO_ORPHAN_PREFIX_LEN + 1, "%s",
2142                                 OCFS2_DIO_ORPHAN_PREFIX);
2143                 if (ret != OCFS2_DIO_ORPHAN_PREFIX_LEN) {
2144                         ret = -EINVAL;
2145                         mlog_errno(ret);
2146                         return ret;
2147                 }
2148
2149                 ret = ocfs2_blkno_stringify(blkno,
2150                                 name + OCFS2_DIO_ORPHAN_PREFIX_LEN);
2151         } else
2152                 ret = ocfs2_blkno_stringify(blkno, name);
2153         if (ret < 0) {
2154                 mlog_errno(ret);
2155                 return ret;
2156         }
2157
2158         ret = ocfs2_prepare_dir_for_insert(osb, orphan_dir_inode,
2159                                            orphan_dir_bh, name,
2160                                            namelen, lookup);
2161         if (ret < 0) {
2162                 mlog_errno(ret);
2163                 return ret;
2164         }
2165
2166         return 0;
2167 }
2168
2169 /**
2170  * ocfs2_prepare_orphan_dir() - Prepare an orphan directory for
2171  * insertion of an orphan.
2172  * @osb: ocfs2 file system
2173  * @ret_orphan_dir: Orphan dir inode - returned locked!
2174  * @blkno: Actual block number of the inode to be inserted into orphan dir.
2175  * @lookup: dir lookup result, to be passed back into functions like
2176  *          ocfs2_orphan_add
2177  *
2178  * Returns zero on success and the ret_orphan_dir, name and lookup
2179  * fields will be populated.
2180  *
2181  * Returns non-zero on failure.
2182  */
2183 static int ocfs2_prepare_orphan_dir(struct ocfs2_super *osb,
2184                                     struct inode **ret_orphan_dir,
2185                                     u64 blkno,
2186                                     char *name,
2187                                     struct ocfs2_dir_lookup_result *lookup,
2188                                     bool dio)
2189 {
2190         struct inode *orphan_dir_inode = NULL;
2191         struct buffer_head *orphan_dir_bh = NULL;
2192         int ret = 0;
2193
2194         ret = ocfs2_lookup_lock_orphan_dir(osb, &orphan_dir_inode,
2195                                            &orphan_dir_bh);
2196         if (ret < 0) {
2197                 mlog_errno(ret);
2198                 return ret;
2199         }
2200
2201         ret = __ocfs2_prepare_orphan_dir(orphan_dir_inode, orphan_dir_bh,
2202                                          blkno, name, lookup, dio);
2203         if (ret < 0) {
2204                 mlog_errno(ret);
2205                 goto out;
2206         }
2207
2208         *ret_orphan_dir = orphan_dir_inode;
2209
2210 out:
2211         brelse(orphan_dir_bh);
2212
2213         if (ret) {
2214                 ocfs2_inode_unlock(orphan_dir_inode, 1);
2215                 inode_unlock(orphan_dir_inode);
2216                 iput(orphan_dir_inode);
2217         }
2218
2219         if (ret)
2220                 mlog_errno(ret);
2221         return ret;
2222 }
2223
2224 static int ocfs2_orphan_add(struct ocfs2_super *osb,
2225                             handle_t *handle,
2226                             struct inode *inode,
2227                             struct buffer_head *fe_bh,
2228                             char *name,
2229                             struct ocfs2_dir_lookup_result *lookup,
2230                             struct inode *orphan_dir_inode,
2231                             bool dio)
2232 {
2233         struct buffer_head *orphan_dir_bh = NULL;
2234         int status = 0;
2235         struct ocfs2_dinode *orphan_fe;
2236         struct ocfs2_dinode *fe = (struct ocfs2_dinode *) fe_bh->b_data;
2237         int namelen = dio ?
2238                         (OCFS2_DIO_ORPHAN_PREFIX_LEN + OCFS2_ORPHAN_NAMELEN) :
2239                         OCFS2_ORPHAN_NAMELEN;
2240
2241         trace_ocfs2_orphan_add_begin(
2242                                 (unsigned long long)OCFS2_I(inode)->ip_blkno);
2243
2244         status = ocfs2_read_inode_block(orphan_dir_inode, &orphan_dir_bh);
2245         if (status < 0) {
2246                 mlog_errno(status);
2247                 goto leave;
2248         }
2249
2250         status = ocfs2_journal_access_di(handle,
2251                                          INODE_CACHE(orphan_dir_inode),
2252                                          orphan_dir_bh,
2253                                          OCFS2_JOURNAL_ACCESS_WRITE);
2254         if (status < 0) {
2255                 mlog_errno(status);
2256                 goto leave;
2257         }
2258
2259         /*
2260          * We're going to journal the change of i_flags and i_orphaned_slot.
2261          * It's safe anyway, though some callers may duplicate the journaling.
2262          * Journaling within the func just make the logic look more
2263          * straightforward.
2264          */
2265         status = ocfs2_journal_access_di(handle,
2266                                          INODE_CACHE(inode),
2267                                          fe_bh,
2268                                          OCFS2_JOURNAL_ACCESS_WRITE);
2269         if (status < 0) {
2270                 mlog_errno(status);
2271                 goto leave;
2272         }
2273
2274         /* we're a cluster, and nlink can change on disk from
2275          * underneath us... */
2276         orphan_fe = (struct ocfs2_dinode *) orphan_dir_bh->b_data;
2277         if (S_ISDIR(inode->i_mode))
2278                 ocfs2_add_links_count(orphan_fe, 1);
2279         set_nlink(orphan_dir_inode, ocfs2_read_links_count(orphan_fe));
2280         ocfs2_journal_dirty(handle, orphan_dir_bh);
2281
2282         status = __ocfs2_add_entry(handle, orphan_dir_inode, name,
2283                                    namelen, inode,
2284                                    OCFS2_I(inode)->ip_blkno,
2285                                    orphan_dir_bh, lookup);
2286         if (status < 0) {
2287                 mlog_errno(status);
2288                 goto rollback;
2289         }
2290
2291         if (dio) {
2292                 /* Update flag OCFS2_DIO_ORPHANED_FL and record the orphan
2293                  * slot.
2294                  */
2295                 fe->i_flags |= cpu_to_le32(OCFS2_DIO_ORPHANED_FL);
2296                 fe->i_dio_orphaned_slot = cpu_to_le16(osb->slot_num);
2297         } else {
2298                 fe->i_flags |= cpu_to_le32(OCFS2_ORPHANED_FL);
2299                 OCFS2_I(inode)->ip_flags &= ~OCFS2_INODE_SKIP_ORPHAN_DIR;
2300
2301                 /* Record which orphan dir our inode now resides
2302                  * in. delete_inode will use this to determine which orphan
2303                  * dir to lock. */
2304                 fe->i_orphaned_slot = cpu_to_le16(osb->slot_num);
2305         }
2306
2307         ocfs2_journal_dirty(handle, fe_bh);
2308
2309         trace_ocfs2_orphan_add_end((unsigned long long)OCFS2_I(inode)->ip_blkno,
2310                                    osb->slot_num);
2311
2312 rollback:
2313         if (status < 0) {
2314                 if (S_ISDIR(inode->i_mode))
2315                         ocfs2_add_links_count(orphan_fe, -1);
2316                 set_nlink(orphan_dir_inode, ocfs2_read_links_count(orphan_fe));
2317         }
2318
2319 leave:
2320         brelse(orphan_dir_bh);
2321
2322         return status;
2323 }
2324
2325 /* unlike orphan_add, we expect the orphan dir to already be locked here. */
2326 int ocfs2_orphan_del(struct ocfs2_super *osb,
2327                      handle_t *handle,
2328                      struct inode *orphan_dir_inode,
2329                      struct inode *inode,
2330                      struct buffer_head *orphan_dir_bh,
2331                      bool dio)
2332 {
2333         char name[OCFS2_DIO_ORPHAN_PREFIX_LEN + OCFS2_ORPHAN_NAMELEN + 1];
2334         struct ocfs2_dinode *orphan_fe;
2335         int status = 0;
2336         struct ocfs2_dir_lookup_result lookup = { NULL, };
2337
2338         if (dio) {
2339                 status = snprintf(name, OCFS2_DIO_ORPHAN_PREFIX_LEN + 1, "%s",
2340                                 OCFS2_DIO_ORPHAN_PREFIX);
2341                 if (status != OCFS2_DIO_ORPHAN_PREFIX_LEN) {
2342                         status = -EINVAL;
2343                         mlog_errno(status);
2344                         return status;
2345                 }
2346
2347                 status = ocfs2_blkno_stringify(OCFS2_I(inode)->ip_blkno,
2348                                 name + OCFS2_DIO_ORPHAN_PREFIX_LEN);
2349         } else
2350                 status = ocfs2_blkno_stringify(OCFS2_I(inode)->ip_blkno, name);
2351         if (status < 0) {
2352                 mlog_errno(status);
2353                 goto leave;
2354         }
2355
2356         trace_ocfs2_orphan_del(
2357              (unsigned long long)OCFS2_I(orphan_dir_inode)->ip_blkno,
2358              name, strlen(name));
2359
2360         status = ocfs2_journal_access_di(handle,
2361                                          INODE_CACHE(orphan_dir_inode),
2362                                          orphan_dir_bh,
2363                                          OCFS2_JOURNAL_ACCESS_WRITE);
2364         if (status < 0) {
2365                 mlog_errno(status);
2366                 goto leave;
2367         }
2368
2369         /* find it's spot in the orphan directory */
2370         status = ocfs2_find_entry(name, strlen(name), orphan_dir_inode,
2371                                   &lookup);
2372         if (status) {
2373                 mlog_errno(status);
2374                 goto leave;
2375         }
2376
2377         /* remove it from the orphan directory */
2378         status = ocfs2_delete_entry(handle, orphan_dir_inode, &lookup);
2379         if (status < 0) {
2380                 mlog_errno(status);
2381                 goto leave;
2382         }
2383
2384         /* do the i_nlink dance! :) */
2385         orphan_fe = (struct ocfs2_dinode *) orphan_dir_bh->b_data;
2386         if (S_ISDIR(inode->i_mode))
2387                 ocfs2_add_links_count(orphan_fe, -1);
2388         set_nlink(orphan_dir_inode, ocfs2_read_links_count(orphan_fe));
2389         ocfs2_journal_dirty(handle, orphan_dir_bh);
2390
2391 leave:
2392         ocfs2_free_dir_lookup_result(&lookup);
2393
2394         if (status)
2395                 mlog_errno(status);
2396         return status;
2397 }
2398
2399 /**
2400  * ocfs2_prep_new_orphaned_file() - Prepare the orphan dir to receive a newly
2401  * allocated file. This is different from the typical 'add to orphan dir'
2402  * operation in that the inode does not yet exist. This is a problem because
2403  * the orphan dir stringifies the inode block number to come up with it's
2404  * dirent. Obviously if the inode does not yet exist we have a chicken and egg
2405  * problem. This function works around it by calling deeper into the orphan
2406  * and suballoc code than other callers. Use this only by necessity.
2407  * @dir: The directory which this inode will ultimately wind up under - not the
2408  * orphan dir!
2409  * @dir_bh: buffer_head the @dir inode block
2410  * @orphan_name: string of length (CFS2_ORPHAN_NAMELEN + 1). Will be filled
2411  * with the string to be used for orphan dirent. Pass back to the orphan dir
2412  * code.
2413  * @ret_orphan_dir: orphan dir inode returned to be passed back into orphan
2414  * dir code.
2415  * @ret_di_blkno: block number where the new inode will be allocated.
2416  * @orphan_insert: Dir insert context to be passed back into orphan dir code.
2417  * @ret_inode_ac: Inode alloc context to be passed back to the allocator.
2418  *
2419  * Returns zero on success and the ret_orphan_dir, name and lookup
2420  * fields will be populated.
2421  *
2422  * Returns non-zero on failure.
2423  */
2424 static int ocfs2_prep_new_orphaned_file(struct inode *dir,
2425                                         struct buffer_head *dir_bh,
2426                                         char *orphan_name,
2427                                         struct inode **ret_orphan_dir,
2428                                         u64 *ret_di_blkno,
2429                                         struct ocfs2_dir_lookup_result *orphan_insert,
2430                                         struct ocfs2_alloc_context **ret_inode_ac)
2431 {
2432         int ret;
2433         u64 di_blkno;
2434         struct ocfs2_super *osb = OCFS2_SB(dir->i_sb);
2435         struct inode *orphan_dir = NULL;
2436         struct buffer_head *orphan_dir_bh = NULL;
2437         struct ocfs2_alloc_context *inode_ac = NULL;
2438
2439         ret = ocfs2_lookup_lock_orphan_dir(osb, &orphan_dir, &orphan_dir_bh);
2440         if (ret < 0) {
2441                 mlog_errno(ret);
2442                 return ret;
2443         }
2444
2445         /* reserve an inode spot */
2446         ret = ocfs2_reserve_new_inode(osb, &inode_ac);
2447         if (ret < 0) {
2448                 if (ret != -ENOSPC)
2449                         mlog_errno(ret);
2450                 goto out;
2451         }
2452
2453         ret = ocfs2_find_new_inode_loc(dir, dir_bh, inode_ac,
2454                                        &di_blkno);
2455         if (ret) {
2456                 mlog_errno(ret);
2457                 goto out;
2458         }
2459
2460         ret = __ocfs2_prepare_orphan_dir(orphan_dir, orphan_dir_bh,
2461                                          di_blkno, orphan_name, orphan_insert,
2462                                          false);
2463         if (ret < 0) {
2464                 mlog_errno(ret);
2465                 goto out;
2466         }
2467
2468 out:
2469         if (ret == 0) {
2470                 *ret_orphan_dir = orphan_dir;
2471                 *ret_di_blkno = di_blkno;
2472                 *ret_inode_ac = inode_ac;
2473                 /*
2474                  * orphan_name and orphan_insert are already up to
2475                  * date via prepare_orphan_dir
2476                  */
2477         } else {
2478                 /* Unroll reserve_new_inode* */
2479                 if (inode_ac)
2480                         ocfs2_free_alloc_context(inode_ac);
2481
2482                 /* Unroll orphan dir locking */
2483                 inode_unlock(orphan_dir);
2484                 ocfs2_inode_unlock(orphan_dir, 1);
2485                 iput(orphan_dir);
2486         }
2487
2488         brelse(orphan_dir_bh);
2489
2490         return ret;
2491 }
2492
2493 int ocfs2_create_inode_in_orphan(struct inode *dir,
2494                                  int mode,
2495                                  struct inode **new_inode)
2496 {
2497         int status, did_quota_inode = 0;
2498         struct inode *inode = NULL;
2499         struct inode *orphan_dir = NULL;
2500         struct ocfs2_super *osb = OCFS2_SB(dir->i_sb);
2501         handle_t *handle = NULL;
2502         char orphan_name[OCFS2_ORPHAN_NAMELEN + 1];
2503         struct buffer_head *parent_di_bh = NULL;
2504         struct buffer_head *new_di_bh = NULL;
2505         struct ocfs2_alloc_context *inode_ac = NULL;
2506         struct ocfs2_dir_lookup_result orphan_insert = { NULL, };
2507         u64 di_blkno, suballoc_loc;
2508         u16 suballoc_bit;
2509
2510         status = ocfs2_inode_lock(dir, &parent_di_bh, 1);
2511         if (status < 0) {
2512                 if (status != -ENOENT)
2513                         mlog_errno(status);
2514                 return status;
2515         }
2516
2517         status = ocfs2_prep_new_orphaned_file(dir, parent_di_bh,
2518                                               orphan_name, &orphan_dir,
2519                                               &di_blkno, &orphan_insert, &inode_ac);
2520         if (status < 0) {
2521                 if (status != -ENOSPC)
2522                         mlog_errno(status);
2523                 goto leave;
2524         }
2525
2526         inode = ocfs2_get_init_inode(dir, mode);
2527         if (IS_ERR(inode)) {
2528                 status = PTR_ERR(inode);
2529                 inode = NULL;
2530                 mlog_errno(status);
2531                 goto leave;
2532         }
2533
2534         handle = ocfs2_start_trans(osb, ocfs2_mknod_credits(osb->sb, 0, 0));
2535         if (IS_ERR(handle)) {
2536                 status = PTR_ERR(handle);
2537                 handle = NULL;
2538                 mlog_errno(status);
2539                 goto leave;
2540         }
2541
2542         status = dquot_alloc_inode(inode);
2543         if (status)
2544                 goto leave;
2545         did_quota_inode = 1;
2546
2547         status = ocfs2_claim_new_inode_at_loc(handle, dir, inode_ac,
2548                                               &suballoc_loc,
2549                                               &suballoc_bit, di_blkno);
2550         if (status < 0) {
2551                 mlog_errno(status);
2552                 goto leave;
2553         }
2554
2555         clear_nlink(inode);
2556         /* do the real work now. */
2557         status = __ocfs2_mknod_locked(dir, inode,
2558                                       0, &new_di_bh, parent_di_bh, handle,
2559                                       inode_ac, di_blkno, suballoc_loc,
2560                                       suballoc_bit);
2561         if (status < 0) {
2562                 mlog_errno(status);
2563                 goto leave;
2564         }
2565
2566         status = ocfs2_orphan_add(osb, handle, inode, new_di_bh, orphan_name,
2567                                   &orphan_insert, orphan_dir, false);
2568         if (status < 0) {
2569                 mlog_errno(status);
2570                 goto leave;
2571         }
2572
2573         /* get open lock so that only nodes can't remove it from orphan dir. */
2574         status = ocfs2_open_lock(inode);
2575         if (status < 0)
2576                 mlog_errno(status);
2577
2578         insert_inode_hash(inode);
2579 leave:
2580         if (status < 0 && did_quota_inode)
2581                 dquot_free_inode(inode);
2582         if (handle)
2583                 ocfs2_commit_trans(osb, handle);
2584
2585         if (orphan_dir) {
2586                 /* This was locked for us in ocfs2_prepare_orphan_dir() */
2587                 ocfs2_inode_unlock(orphan_dir, 1);
2588                 inode_unlock(orphan_dir);
2589                 iput(orphan_dir);
2590         }
2591
2592         if ((status < 0) && inode) {
2593                 clear_nlink(inode);
2594                 iput(inode);
2595         }
2596
2597         if (inode_ac)
2598                 ocfs2_free_alloc_context(inode_ac);
2599
2600         brelse(new_di_bh);
2601
2602         if (!status)
2603                 *new_inode = inode;
2604
2605         ocfs2_free_dir_lookup_result(&orphan_insert);
2606
2607         ocfs2_inode_unlock(dir, 1);
2608         brelse(parent_di_bh);
2609         return status;
2610 }
2611
2612 int ocfs2_add_inode_to_orphan(struct ocfs2_super *osb,
2613         struct inode *inode)
2614 {
2615         char orphan_name[OCFS2_DIO_ORPHAN_PREFIX_LEN + OCFS2_ORPHAN_NAMELEN + 1];
2616         struct inode *orphan_dir_inode = NULL;
2617         struct ocfs2_dir_lookup_result orphan_insert = { NULL, };
2618         struct buffer_head *di_bh = NULL;
2619         int status = 0;
2620         handle_t *handle = NULL;
2621         struct ocfs2_dinode *di = NULL;
2622
2623         status = ocfs2_inode_lock(inode, &di_bh, 1);
2624         if (status < 0) {
2625                 mlog_errno(status);
2626                 goto bail;
2627         }
2628
2629         di = (struct ocfs2_dinode *) di_bh->b_data;
2630         /*
2631          * Another append dio crashed?
2632          * If so, manually recover it first.
2633          */
2634         if (unlikely(di->i_flags & cpu_to_le32(OCFS2_DIO_ORPHANED_FL))) {
2635                 status = ocfs2_truncate_file(inode, di_bh, i_size_read(inode));
2636                 if (status < 0) {
2637                         if (status != -ENOSPC)
2638                                 mlog_errno(status);
2639                         goto bail_unlock_inode;
2640                 }
2641
2642                 status = ocfs2_del_inode_from_orphan(osb, inode, di_bh, 0, 0);
2643                 if (status < 0) {
2644                         mlog_errno(status);
2645                         goto bail_unlock_inode;
2646                 }
2647         }
2648
2649         status = ocfs2_prepare_orphan_dir(osb, &orphan_dir_inode,
2650                         OCFS2_I(inode)->ip_blkno,
2651                         orphan_name,
2652                         &orphan_insert,
2653                         true);
2654         if (status < 0) {
2655                 mlog_errno(status);
2656                 goto bail_unlock_inode;
2657         }
2658
2659         handle = ocfs2_start_trans(osb,
2660                         OCFS2_INODE_ADD_TO_ORPHAN_CREDITS);
2661         if (IS_ERR(handle)) {
2662                 status = PTR_ERR(handle);
2663                 goto bail_unlock_orphan;
2664         }
2665
2666         status = ocfs2_orphan_add(osb, handle, inode, di_bh, orphan_name,
2667                         &orphan_insert, orphan_dir_inode, true);
2668         if (status)
2669                 mlog_errno(status);
2670
2671         ocfs2_commit_trans(osb, handle);
2672
2673 bail_unlock_orphan:
2674         ocfs2_inode_unlock(orphan_dir_inode, 1);
2675         inode_unlock(orphan_dir_inode);
2676         iput(orphan_dir_inode);
2677
2678         ocfs2_free_dir_lookup_result(&orphan_insert);
2679
2680 bail_unlock_inode:
2681         ocfs2_inode_unlock(inode, 1);
2682         brelse(di_bh);
2683
2684 bail:
2685         return status;
2686 }
2687
2688 int ocfs2_del_inode_from_orphan(struct ocfs2_super *osb,
2689                 struct inode *inode, struct buffer_head *di_bh,
2690                 int update_isize, loff_t end)
2691 {
2692         struct inode *orphan_dir_inode = NULL;
2693         struct buffer_head *orphan_dir_bh = NULL;
2694         struct ocfs2_dinode *di = (struct ocfs2_dinode *)di_bh->b_data;
2695         handle_t *handle = NULL;
2696         int status = 0;
2697
2698         orphan_dir_inode = ocfs2_get_system_file_inode(osb,
2699                         ORPHAN_DIR_SYSTEM_INODE,
2700                         le16_to_cpu(di->i_dio_orphaned_slot));
2701         if (!orphan_dir_inode) {
2702                 status = -ENOENT;
2703                 mlog_errno(status);
2704                 goto bail;
2705         }
2706
2707         inode_lock(orphan_dir_inode);
2708         status = ocfs2_inode_lock(orphan_dir_inode, &orphan_dir_bh, 1);
2709         if (status < 0) {
2710                 inode_unlock(orphan_dir_inode);
2711                 iput(orphan_dir_inode);
2712                 mlog_errno(status);
2713                 goto bail;
2714         }
2715
2716         handle = ocfs2_start_trans(osb,
2717                         OCFS2_INODE_DEL_FROM_ORPHAN_CREDITS);
2718         if (IS_ERR(handle)) {
2719                 status = PTR_ERR(handle);
2720                 goto bail_unlock_orphan;
2721         }
2722
2723         BUG_ON(!(di->i_flags & cpu_to_le32(OCFS2_DIO_ORPHANED_FL)));
2724
2725         status = ocfs2_orphan_del(osb, handle, orphan_dir_inode,
2726                                 inode, orphan_dir_bh, true);
2727         if (status < 0) {
2728                 mlog_errno(status);
2729                 goto bail_commit;
2730         }
2731
2732         status = ocfs2_journal_access_di(handle,
2733                         INODE_CACHE(inode),
2734                         di_bh,
2735                         OCFS2_JOURNAL_ACCESS_WRITE);
2736         if (status < 0) {
2737                 mlog_errno(status);
2738                 goto bail_commit;
2739         }
2740
2741         di->i_flags &= ~cpu_to_le32(OCFS2_DIO_ORPHANED_FL);
2742         di->i_dio_orphaned_slot = 0;
2743
2744         if (update_isize) {
2745                 status = ocfs2_set_inode_size(handle, inode, di_bh, end);
2746                 if (status)
2747                         mlog_errno(status);
2748         } else
2749                 ocfs2_journal_dirty(handle, di_bh);
2750
2751 bail_commit:
2752         ocfs2_commit_trans(osb, handle);
2753
2754 bail_unlock_orphan:
2755         ocfs2_inode_unlock(orphan_dir_inode, 1);
2756         inode_unlock(orphan_dir_inode);
2757         brelse(orphan_dir_bh);
2758         iput(orphan_dir_inode);
2759
2760 bail:
2761         return status;
2762 }
2763
2764 int ocfs2_mv_orphaned_inode_to_new(struct inode *dir,
2765                                    struct inode *inode,
2766                                    struct dentry *dentry)
2767 {
2768         int status = 0;
2769         struct buffer_head *parent_di_bh = NULL;
2770         handle_t *handle = NULL;
2771         struct ocfs2_super *osb = OCFS2_SB(dir->i_sb);
2772         struct ocfs2_dinode *dir_di, *di;
2773         struct inode *orphan_dir_inode = NULL;
2774         struct buffer_head *orphan_dir_bh = NULL;
2775         struct buffer_head *di_bh = NULL;
2776         struct ocfs2_dir_lookup_result lookup = { NULL, };
2777
2778         trace_ocfs2_mv_orphaned_inode_to_new(dir, dentry,
2779                                 dentry->d_name.len, dentry->d_name.name,
2780                                 (unsigned long long)OCFS2_I(dir)->ip_blkno,
2781                                 (unsigned long long)OCFS2_I(inode)->ip_blkno);
2782
2783         status = ocfs2_inode_lock(dir, &parent_di_bh, 1);
2784         if (status < 0) {
2785                 if (status != -ENOENT)
2786                         mlog_errno(status);
2787                 return status;
2788         }
2789
2790         dir_di = (struct ocfs2_dinode *) parent_di_bh->b_data;
2791         if (!dir_di->i_links_count) {
2792                 /* can't make a file in a deleted directory. */
2793                 status = -ENOENT;
2794                 goto leave;
2795         }
2796
2797         status = ocfs2_check_dir_for_entry(dir, dentry->d_name.name,
2798                                            dentry->d_name.len);
2799         if (status)
2800                 goto leave;
2801
2802         /* get a spot inside the dir. */
2803         status = ocfs2_prepare_dir_for_insert(osb, dir, parent_di_bh,
2804                                               dentry->d_name.name,
2805                                               dentry->d_name.len, &lookup);
2806         if (status < 0) {
2807                 mlog_errno(status);
2808                 goto leave;
2809         }
2810
2811         orphan_dir_inode = ocfs2_get_system_file_inode(osb,
2812                                                        ORPHAN_DIR_SYSTEM_INODE,
2813                                                        osb->slot_num);
2814         if (!orphan_dir_inode) {
2815                 status = -ENOENT;
2816                 mlog_errno(status);
2817                 goto leave;
2818         }
2819
2820         inode_lock(orphan_dir_inode);
2821
2822         status = ocfs2_inode_lock(orphan_dir_inode, &orphan_dir_bh, 1);
2823         if (status < 0) {
2824                 mlog_errno(status);
2825                 inode_unlock(orphan_dir_inode);
2826                 iput(orphan_dir_inode);
2827                 goto leave;
2828         }
2829
2830         status = ocfs2_read_inode_block(inode, &di_bh);
2831         if (status < 0) {
2832                 mlog_errno(status);
2833                 goto orphan_unlock;
2834         }
2835
2836         handle = ocfs2_start_trans(osb, ocfs2_rename_credits(osb->sb));
2837         if (IS_ERR(handle)) {
2838                 status = PTR_ERR(handle);
2839                 handle = NULL;
2840                 mlog_errno(status);
2841                 goto orphan_unlock;
2842         }
2843
2844         status = ocfs2_journal_access_di(handle, INODE_CACHE(inode),
2845                                          di_bh, OCFS2_JOURNAL_ACCESS_WRITE);
2846         if (status < 0) {
2847                 mlog_errno(status);
2848                 goto out_commit;
2849         }
2850
2851         status = ocfs2_orphan_del(osb, handle, orphan_dir_inode, inode,
2852                                   orphan_dir_bh, false);
2853         if (status < 0) {
2854                 mlog_errno(status);
2855                 goto out_commit;
2856         }
2857
2858         di = (struct ocfs2_dinode *)di_bh->b_data;
2859         di->i_flags &= ~cpu_to_le32(OCFS2_ORPHANED_FL);
2860         di->i_orphaned_slot = 0;
2861         set_nlink(inode, 1);
2862         ocfs2_set_links_count(di, inode->i_nlink);
2863         ocfs2_update_inode_fsync_trans(handle, inode, 1);
2864         ocfs2_journal_dirty(handle, di_bh);
2865
2866         status = ocfs2_add_entry(handle, dentry, inode,
2867                                  OCFS2_I(inode)->ip_blkno, parent_di_bh,
2868                                  &lookup);
2869         if (status < 0) {
2870                 mlog_errno(status);
2871                 goto out_commit;
2872         }
2873
2874         status = ocfs2_dentry_attach_lock(dentry, inode,
2875                                           OCFS2_I(dir)->ip_blkno);
2876         if (status) {
2877                 mlog_errno(status);
2878                 goto out_commit;
2879         }
2880
2881         d_instantiate(dentry, inode);
2882         status = 0;
2883 out_commit:
2884         ocfs2_commit_trans(osb, handle);
2885 orphan_unlock:
2886         ocfs2_inode_unlock(orphan_dir_inode, 1);
2887         inode_unlock(orphan_dir_inode);
2888         iput(orphan_dir_inode);
2889 leave:
2890
2891         ocfs2_inode_unlock(dir, 1);
2892
2893         brelse(di_bh);
2894         brelse(parent_di_bh);
2895         brelse(orphan_dir_bh);
2896
2897         ocfs2_free_dir_lookup_result(&lookup);
2898
2899         if (status)
2900                 mlog_errno(status);
2901
2902         return status;
2903 }
2904
2905 const struct inode_operations ocfs2_dir_iops = {
2906         .create         = ocfs2_create,
2907         .lookup         = ocfs2_lookup,
2908         .link           = ocfs2_link,
2909         .unlink         = ocfs2_unlink,
2910         .rmdir          = ocfs2_unlink,
2911         .symlink        = ocfs2_symlink,
2912         .mkdir          = ocfs2_mkdir,
2913         .mknod          = ocfs2_mknod,
2914         .rename         = ocfs2_rename,
2915         .setattr        = ocfs2_setattr,
2916         .getattr        = ocfs2_getattr,
2917         .permission     = ocfs2_permission,
2918         .listxattr      = ocfs2_listxattr,
2919         .fiemap         = ocfs2_fiemap,
2920         .get_acl        = ocfs2_iop_get_acl,
2921         .set_acl        = ocfs2_iop_set_acl,
2922 };